diff --git a/paddle/fluid/eager/accumulation/accumulation_node.cc b/paddle/fluid/eager/accumulation/accumulation_node.cc
index ed1146eed0fb02df6e5c01381f2224d0b0d069b8..823c0153d71b029235ace7a262d8440b339b89f9 100644
--- a/paddle/fluid/eager/accumulation/accumulation_node.cc
+++ b/paddle/fluid/eager/accumulation/accumulation_node.cc
@@ -18,7 +18,6 @@
 
 #include "paddle/pten/api/all.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
diff --git a/paddle/fluid/eager/accumulation/gradient_accumulation.cc b/paddle/fluid/eager/accumulation/gradient_accumulation.cc
index 9bc24dd28756a54eefcdc4f75c0450f073c257ae..1f66596a0b5783b537f3674b376f88f59a29fc87 100644
--- a/paddle/fluid/eager/accumulation/gradient_accumulation.cc
+++ b/paddle/fluid/eager/accumulation/gradient_accumulation.cc
@@ -28,7 +28,6 @@
 #include "paddle/fluid/platform/float16.h"
 #include "paddle/pten/api/all.h"
 #include "paddle/pten/core/convert_utils.h"
-#include "paddle/pten/include/core.h"
 #include "unsupported/Eigen/CXX11/Tensor"
 #ifdef PADDLE_WITH_XPU
 #include "xpu/refactor/math.h"
diff --git a/paddle/fluid/eager/api/generated/eager_generated/backwards/scale_node.cc b/paddle/fluid/eager/api/generated/eager_generated/backwards/scale_node.cc
index 02eaa79fc9b28aa689e38c22075f535ad47387a9..99f6c7a83538e93d8ac27801c665cc6b44f307fc 100644
--- a/paddle/fluid/eager/api/generated/eager_generated/backwards/scale_node.cc
+++ b/paddle/fluid/eager/api/generated/eager_generated/backwards/scale_node.cc
@@ -16,7 +16,7 @@
 #include "paddle/fluid/eager/api/utils/global_utils.h"
 #include "paddle/fluid/eager/eager_tensor.h"
 
-#include "paddle/pten/api/all.h"
+#include "paddle/pten/kernels/scale_kernel.h"
 
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
@@ -33,28 +33,28 @@ static void ScaleDeviceDispatch(const pten::DenseTensor& dense_tensor,
                                 pten::DenseTensor* dense_out) {
   switch (dense_tensor.dtype()) {
     case pten::DataType::FLOAT64: {
-      pten::Scale<double, DeviceContext>(
+      pten::ScaleKernel<double, DeviceContext>(
           dev_ctx, dense_tensor /* tensor */, scale /* scale */,
           bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);
       break;
     }
     case pten::DataType::FLOAT32: {
-      pten::Scale<float, DeviceContext>(dev_ctx, dense_tensor /* tensor */,
-                                        scale /* scale */, bias /* bias */,
-                                        bias_after_scale /* bias_after_scale */,
-                                        dense_out /* out tensor */);
+      pten::ScaleKernel<float, DeviceContext>(
+          dev_ctx, dense_tensor /* tensor */, scale /* scale */,
+          bias /* bias */, bias_after_scale /* bias_after_scale */,
+          dense_out /* out tensor */);
       break;
     }
     case pten::DataType::INT64: {
-      pten::Scale<int64_t, DeviceContext>(
+      pten::ScaleKernel<int64_t, DeviceContext>(
           dev_ctx, dense_tensor /* tensor */, scale /* scale */,
           bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);
       break;
     }
     case pten::DataType::INT32: {
-      pten::Scale<int32_t, DeviceContext>(
+      pten::ScaleKernel<int32_t, DeviceContext>(
           dev_ctx, dense_tensor /* tensor */, scale /* scale */,
           bias /* bias */, bias_after_scale /* bias_after_scale */,
           dense_out /* out tensor */);
diff --git a/paddle/fluid/eager/api/generated/eager_generated/forwards/scale.cc b/paddle/fluid/eager/api/generated/eager_generated/forwards/scale.cc
index 7b20ff144a7a76aa85163c6ea4e1250dbf1118de..642302a4119bec432a6b07d8dca5021a472ab131 100644
--- a/paddle/fluid/eager/api/generated/eager_generated/forwards/scale.cc
+++ b/paddle/fluid/eager/api/generated/eager_generated/forwards/scale.cc
@@ -29,7 +29,6 @@
 #include "paddle/fluid/eager/utils.h"
 
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 
 namespace egr {
 
diff --git a/paddle/fluid/eager/eager_tensor.h b/paddle/fluid/eager/eager_tensor.h
index 80faad9080ffe0c3daf33ed28ed2e8000372f9b5..c58c0b9e66e7aee43e311e54074f192ec1bcfae6 100644
--- a/paddle/fluid/eager/eager_tensor.h
+++ b/paddle/fluid/eager/eager_tensor.h
@@ -18,7 +18,6 @@
 #include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/framework/variable.h"
 // pten deps
-#include "paddle/pten/all.h"
 #include "paddle/pten/api/all.h"
 #include "paddle/pten/api/lib/api_declare.h"
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
diff --git a/paddle/fluid/eager/grad_node_info.h b/paddle/fluid/eager/grad_node_info.h
index f15c50ef75190da94316acbb5d06afa5f54fc02d..5cf0b90220148fdf91b056e2f7900555ca91bfad 100644
--- a/paddle/fluid/eager/grad_node_info.h
+++ b/paddle/fluid/eager/grad_node_info.h
@@ -16,7 +16,6 @@
 
 #include "paddle/fluid/eager/eager_tensor.h"
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 
 namespace egr {
 /**
diff --git a/paddle/fluid/eager/legacy/infer_var_type_context.h b/paddle/fluid/eager/legacy/infer_var_type_context.h
index 2d5a8d806fee7c8ad9426aa0503a65b2c0d60e92..9d9cbeb38ccfabd7cd30714f7253d8cd863ac308 100644
--- a/paddle/fluid/eager/legacy/infer_var_type_context.h
+++ b/paddle/fluid/eager/legacy/infer_var_type_context.h
@@ -26,7 +26,6 @@
 #include "paddle/fluid/framework/var_type_inference.h"
 #include "paddle/fluid/framework/var_type_traits.h"
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 
 namespace egr {
 namespace legacy {
diff --git a/paddle/fluid/eager/legacy/prepared_operator.h b/paddle/fluid/eager/legacy/prepared_operator.h
index 9ba186b14e3b31a01f8f9814a674016dadbdc62d..0e00b52e0481aaccc94c9f9d1347e3944c092aba 100644
--- a/paddle/fluid/eager/legacy/prepared_operator.h
+++ b/paddle/fluid/eager/legacy/prepared_operator.h
@@ -25,8 +25,6 @@
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/type_defs.h"
 
-#include "paddle/pten/include/core.h"
-
 DECLARE_bool(use_mkldnn);
 
 namespace paddle {
diff --git a/paddle/fluid/eager/legacy/tensor_helper.h b/paddle/fluid/eager/legacy/tensor_helper.h
index f87ab70c93686ecca4b65147c2a25ce52601bbe0..ce407f8965aa0cb2e860e945f14007dde84d0486 100644
--- a/paddle/fluid/eager/legacy/tensor_helper.h
+++ b/paddle/fluid/eager/legacy/tensor_helper.h
@@ -17,7 +17,6 @@
 #include <vector>
 #include "paddle/fluid/eager/eager_tensor.h"
 #include "paddle/pten/api/all.h"
-#include "paddle/pten/include/core.h"
 namespace egr {
 namespace legacy {
 
diff --git a/paddle/fluid/framework/data_device_transform_test.cu b/paddle/fluid/framework/data_device_transform_test.cu
index a81e4abd45e56f3e13bcfe5585ad8906dccd8bdb..858688dffd8c1cf6716662d71e01279ca1386254 100644
--- a/paddle/fluid/framework/data_device_transform_test.cu
+++ b/paddle/fluid/framework/data_device_transform_test.cu
@@ -24,7 +24,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/init.h"
 
 #include "paddle/fluid/framework/pten_utils.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace framework {
diff --git a/paddle/fluid/framework/operator.h b/paddle/fluid/framework/operator.h
index 8e69f96dfb813446033f5dc2bbcd0c1f4de3e4a3..9d75c66beb7d408881d143d1655c1e463796c10d 100644
--- a/paddle/fluid/framework/operator.h
+++ b/paddle/fluid/framework/operator.h
@@ -41,7 +41,8 @@ limitations under the License. */
 #include "paddle/utils/flat_hash_map.h"
 
 #include "paddle/pten/core/arg_map_context.h"
-#include "paddle/pten/include/core.h"
+#include "paddle/pten/core/kernel_context.h"
+#include "paddle/pten/core/kernel_factory.h"
 
 namespace paddle {
 namespace framework {
diff --git a/paddle/fluid/imperative/layer.h b/paddle/fluid/imperative/layer.h
index 199d62bff1f20d5822f9304d46d0389024dc60b0..d27460aeeccef9703d98a5b4d7a61dad91e1f545 100644
--- a/paddle/fluid/imperative/layer.h
+++ b/paddle/fluid/imperative/layer.h
@@ -37,7 +37,6 @@
 #include "paddle/fluid/imperative/variable_wrapper.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/macros.h"
-#include "paddle/pten/include/core.h"
 namespace paddle {
 namespace framework {
 class Variable;
diff --git a/paddle/fluid/imperative/op_base.h b/paddle/fluid/imperative/op_base.h
index 3ff451f81787209eca36acdf9aac62519a90f399..cb76a823532828a3b966638671c67258931493c7 100644
--- a/paddle/fluid/imperative/op_base.h
+++ b/paddle/fluid/imperative/op_base.h
@@ -25,7 +25,6 @@
 #include "paddle/fluid/imperative/type_defs.h"
 #include "paddle/fluid/imperative/variable_wrapper.h"
 #include "paddle/fluid/platform/place.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace imperative {
diff --git a/paddle/fluid/imperative/prepared_operator.h b/paddle/fluid/imperative/prepared_operator.h
index 5262b265b1b5397216d6b16abddac8c880acc3f9..29747e79ef6fad9cb714837285c06e2630c0c0be 100644
--- a/paddle/fluid/imperative/prepared_operator.h
+++ b/paddle/fluid/imperative/prepared_operator.h
@@ -27,8 +27,6 @@
 #include "paddle/fluid/imperative/layer.h"
 #include "paddle/fluid/imperative/type_defs.h"
 
-#include "paddle/pten/include/core.h"
-
 DECLARE_bool(use_mkldnn);
 
 namespace paddle {
diff --git a/paddle/fluid/operators/cast_op.h b/paddle/fluid/operators/cast_op.h
index 72aa9a195ec7c961f8e2ddbb4137eaae1051dd6f..c54c811b25b66050586247595c42ca695455d7ef 100644
--- a/paddle/fluid/operators/cast_op.h
+++ b/paddle/fluid/operators/cast_op.h
@@ -19,7 +19,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/transform.h"
 
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/cast_kernel.h"
 
 namespace paddle {
diff --git a/paddle/fluid/operators/conj_op.h b/paddle/fluid/operators/conj_op.h
index 71115c2eba796dc58b863fd71f2b21b657ac52c7..6df982abb8612f18c57b94af9ec9c9b28711ad14 100644
--- a/paddle/fluid/operators/conj_op.h
+++ b/paddle/fluid/operators/conj_op.h
@@ -19,7 +19,6 @@
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/complex_kernel.h"
 
 namespace paddle {
@@ -39,7 +38,7 @@ class ConjKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::ConjKernel<T, DeviceContext>(dev_ctx, *pt_x.get(), pt_out.get());
+    pten::ConjKernel<T>(dev_ctx, *pt_x.get(), pt_out.get());
   }
 };
 
diff --git a/paddle/fluid/operators/dot_op.h b/paddle/fluid/operators/dot_op.h
index 8817e2f3ca79d148d57d5ac8d7176f0c1c8adfa5..ceb8a28e8aa4cc54c384e757bf0099c76eb05cb4 100644
--- a/paddle/fluid/operators/dot_op.h
+++ b/paddle/fluid/operators/dot_op.h
@@ -21,7 +21,6 @@
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/dot_grad_kernel.h"
 #include "paddle/pten/kernels/dot_kernel.h"
 
diff --git a/paddle/fluid/operators/elementwise/elementwise_add_op.h b/paddle/fluid/operators/elementwise/elementwise_add_op.h
index 35807d7c57d47846e9a5cfcb533ce063c40ea4a2..622a6d7edb783c12facd5cb2644a31721e2791d9 100644
--- a/paddle/fluid/operators/elementwise/elementwise_add_op.h
+++ b/paddle/fluid/operators/elementwise/elementwise_add_op.h
@@ -18,7 +18,6 @@ limitations under the License. */
 #include <utility>
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/kernels/math_kernel.h"
 
 namespace paddle {
diff --git a/paddle/fluid/operators/elementwise/elementwise_mul_op.h b/paddle/fluid/operators/elementwise/elementwise_mul_op.h
index 385c7549e07f20cb1502705b729eb9a2a9ceb72f..687340b668a13f43b74cee0f067d84e09e0bc7da 100644
--- a/paddle/fluid/operators/elementwise/elementwise_mul_op.h
+++ b/paddle/fluid/operators/elementwise/elementwise_mul_op.h
@@ -18,7 +18,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 #include "paddle/fluid/platform/cpu_info.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/kernels/math_kernel.h"
 
 namespace paddle {
diff --git a/paddle/fluid/operators/elementwise/elementwise_op_function.h b/paddle/fluid/operators/elementwise/elementwise_op_function.h
index 37d29ed91b3d40a71ed14ed1962ad09e74df0d1b..626046890fb067f0f6455b3438d0d24f273ef4b0 100644
--- a/paddle/fluid/operators/elementwise/elementwise_op_function.h
+++ b/paddle/fluid/operators/elementwise/elementwise_op_function.h
@@ -29,7 +29,6 @@ limitations under the License. */
 #include "paddle/fluid/platform/device/gpu/gpu_info.h"
 #include "paddle/fluid/platform/transform.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
 #include "paddle/pten/kernels/cpu/elementwise.h"
 
diff --git a/paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h b/paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h
index 36ff1ae254d20c445082fa73f6be894acd360ded..9cc741344e50e75cdd5955045af58e797fe079ba 100644
--- a/paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h
+++ b/paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h
@@ -22,7 +22,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/top/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/gpu/elementwise.h"
 
 namespace paddle {
diff --git a/paddle/fluid/operators/elementwise/elementwise_sub_op.h b/paddle/fluid/operators/elementwise/elementwise_sub_op.h
index 09818380d8ea7d274ab05e4dda1399947bd03055..f035e46d1d08270a4967f7ebf101af7ee78a7ffe 100644
--- a/paddle/fluid/operators/elementwise/elementwise_sub_op.h
+++ b/paddle/fluid/operators/elementwise/elementwise_sub_op.h
@@ -16,7 +16,6 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 
-// only can include the headers in paddle/pten/include dirs
 #include "paddle/pten/kernels/math_kernel.h"
 namespace paddle {
 namespace operators {
diff --git a/paddle/fluid/operators/fill_any_like_op.h b/paddle/fluid/operators/fill_any_like_op.h
index 287bbbfa3b34350b1e6f5da381a6a938a4570ecc..19f6e7a4ef51fe9b900d82b879c4d91915ee2694 100644
--- a/paddle/fluid/operators/fill_any_like_op.h
+++ b/paddle/fluid/operators/fill_any_like_op.h
@@ -19,7 +19,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/pten_utils.h"
 
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/full_kernel.h"
 
 namespace paddle {
diff --git a/paddle/fluid/operators/flatten_op.h b/paddle/fluid/operators/flatten_op.h
index ef42619bfe4ff651d13c18c8fbfc203929c356e1..8e54ecb922f5a1d12d2b643e6ab01dd16201d0cf 100644
--- a/paddle/fluid/operators/flatten_op.h
+++ b/paddle/fluid/operators/flatten_op.h
@@ -20,7 +20,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/math/math_function.h"
 #include "paddle/fluid/operators/math/pooling.h"
 #include "paddle/fluid/platform/device_context.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 #include "paddle/pten/kernels/flatten_grad_kernel.h"
 #include "paddle/pten/kernels/flatten_kernel.h"
diff --git a/paddle/fluid/operators/matmul_v2_op.h b/paddle/fluid/operators/matmul_v2_op.h
index e93bd212868fd0183bc85d55b75146e9e7ebd1ab..9ab77cdcaec0a7e1c8c323e4ac21111b1ab19170 100644
--- a/paddle/fluid/operators/matmul_v2_op.h
+++ b/paddle/fluid/operators/matmul_v2_op.h
@@ -27,7 +27,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/matmul_grad_kernel.h"
 #include "paddle/pten/kernels/matmul_kernel.h"
 
diff --git a/paddle/fluid/operators/reduce_ops/reduce_op.h b/paddle/fluid/operators/reduce_ops/reduce_op.h
index e1854d8a13d8bc4db6de7f7f77b00916706a1113..eb4d4a5c1680ecc752d709b0a2bc01a47758d888 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_op.h
+++ b/paddle/fluid/operators/reduce_ops/reduce_op.h
@@ -26,8 +26,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/math.h"
 #include "paddle/pten/kernels/cpu/reduce.h"
 
 #if defined(__HIPCC__) || defined(__NVCC__)
diff --git a/paddle/fluid/operators/reshape_op.cc b/paddle/fluid/operators/reshape_op.cc
index a25e53aac5d73d4855a5139965cac34455f63941..47b8da70adbac5541bf9efce010e24ae3215d42a 100644
--- a/paddle/fluid/operators/reshape_op.cc
+++ b/paddle/fluid/operators/reshape_op.cc
@@ -20,7 +20,6 @@ limitations under the License. */
 // only can include the headers in paddle/pten/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
 #include "paddle/pten/common/scalar_array.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/reshape_grad_kernel.h"
 #include "paddle/pten/kernels/reshape_kernel.h"
 namespace paddle {
diff --git a/paddle/fluid/operators/scale_op.h b/paddle/fluid/operators/scale_op.h
index 6011fe9a66b60b8cb3e5438f3233e4f25ca486fb..a6f4f6e27204ed783f9eb9ff86977a208c43c995 100644
--- a/paddle/fluid/operators/scale_op.h
+++ b/paddle/fluid/operators/scale_op.h
@@ -19,7 +19,6 @@ limitations under the License. */
 
 // only can include the headers in paddle/top/api dirs
 #include "paddle/pten/api/lib/utils/tensor_utils.h"
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/scale_kernel.h"
 
 namespace paddle {
@@ -70,8 +69,8 @@ class ScaleKernel : public framework::OpKernel<T> {
     auto pt_out = paddle::experimental::MakePtenDenseTensor(*out);
 
     // call new kernel
-    pten::Scale<T>(dev_ctx, *pt_x.get(), scale, bias, bias_after_scale,
-                   pt_out.get());
+    pten::ScaleKernel<T>(dev_ctx, *pt_x.get(), scale, bias, bias_after_scale,
+                         pt_out.get());
   }
 };
 
diff --git a/paddle/fluid/operators/sign_op.h b/paddle/fluid/operators/sign_op.h
index b8dd44c01b0509026a5f231c606726bcd1effd4c..8294cd2c5f1450b00f225082cf9a3735e0912b20 100644
--- a/paddle/fluid/operators/sign_op.h
+++ b/paddle/fluid/operators/sign_op.h
@@ -19,7 +19,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/pten_utils.h"
 #include "paddle/fluid/operators/eigen/eigen_function.h"
 
-#include "paddle/pten/include/core.h"
 #include "paddle/pten/kernels/sign_kernel.h"
 
 namespace paddle {
diff --git a/paddle/fluid/pybind/eager.cc b/paddle/fluid/pybind/eager.cc
index 9484d506b20fb69c0f2758c3597f6496076c7e30..102bc9f162b0f4ba4a2444c4890eca8635217110 100644
--- a/paddle/fluid/pybind/eager.cc
+++ b/paddle/fluid/pybind/eager.cc
@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 #include "pybind11/numpy.h"
 #include "pybind11/pybind11.h"
 #pragma GCC diagnostic ignored "-Wmissing-field-initializers"
diff --git a/paddle/fluid/pybind/eager_functions.cc b/paddle/fluid/pybind/eager_functions.cc
index 659df6b9b44def9b0c5d4737e96b3de3aca98f73..aaf86bc41aeff043ee0c21cb1e31a5f50a828ac9 100644
--- a/paddle/fluid/pybind/eager_functions.cc
+++ b/paddle/fluid/pybind/eager_functions.cc
@@ -34,7 +34,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace pybind {
diff --git a/paddle/fluid/pybind/eager_method.cc b/paddle/fluid/pybind/eager_method.cc
index a0067f9c64fb157fe675e9ad848b65fce7ba621e..a8c1da2a8b8663f661d6a037afa2c7147ac5ebe7 100644
--- a/paddle/fluid/pybind/eager_method.cc
+++ b/paddle/fluid/pybind/eager_method.cc
@@ -31,7 +31,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 namespace paddle {
 namespace pybind {
 
diff --git a/paddle/fluid/pybind/eager_properties.cc b/paddle/fluid/pybind/eager_properties.cc
index 71b8bbbb1a283be084ae33c87f128cb7f1e05769..038a1254d7ef6e243a03c523f0ed357f63f707ad 100644
--- a/paddle/fluid/pybind/eager_properties.cc
+++ b/paddle/fluid/pybind/eager_properties.cc
@@ -28,7 +28,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 #pragma GCC diagnostic ignored "-Wwrite-strings"
 
 namespace paddle {
diff --git a/paddle/fluid/pybind/eager_utils.cc b/paddle/fluid/pybind/eager_utils.cc
index 9849d0d41611bfbe0c3458dcc08476227e09820d..c1049d240795c3fee3ad7b6d0f394c54f22f0e5e 100644
--- a/paddle/fluid/pybind/eager_utils.cc
+++ b/paddle/fluid/pybind/eager_utils.cc
@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/core.h"
 
 namespace paddle {
 namespace pybind {
diff --git a/paddle/pten/CMakeLists.txt b/paddle/pten/CMakeLists.txt
index 6a823ff3672bf1c4160248caf47a5bf826ee11df..a9b7c7581bc2b1bae022fd2b2f836d86936eca7a 100644
--- a/paddle/pten/CMakeLists.txt
+++ b/paddle/pten/CMakeLists.txt
@@ -29,4 +29,4 @@ get_property(pten_kernels GLOBAL PROPERTY PTEN_KERNELS)
 message(STATUS "All standard pten kernels: ${pten_kernels}")
 set(PTEN_DEPS ${PTEN_DEPS} ${pten_kernels})
 
-cc_library(pten SRCS all.cc DEPS ${PTEN_DEPS})
+cc_library(pten DEPS ${PTEN_DEPS})
diff --git a/paddle/pten/all.cc b/paddle/pten/all.cc
deleted file mode 100644
index d8d96e1cd461edf74c2f29cab61f1fbf4191138b..0000000000000000000000000000000000000000
--- a/paddle/pten/all.cc
+++ /dev/null
@@ -1,17 +0,0 @@
-/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#include "paddle/pten/all.h"
-
-namespace pten {}  // namespace pten
diff --git a/paddle/pten/all.h b/paddle/pten/all.h
deleted file mode 100644
index c8be629b10e75ee789795ec1fbeea1d47c4a5166..0000000000000000000000000000000000000000
--- a/paddle/pten/all.h
+++ /dev/null
@@ -1,20 +0,0 @@
-/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#pragma once
-
-// developer apis
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/include/math.h"
diff --git a/paddle/pten/api/lib/utils.cc b/paddle/pten/api/lib/utils.cc
index ddb29c8833f3b8c7f54d9e15acc17d91fdced52d..6eb1e5a3797c916defa7ec502503eab313172ffe 100644
--- a/paddle/pten/api/lib/utils.cc
+++ b/paddle/pten/api/lib/utils.cc
@@ -22,8 +22,7 @@ limitations under the License. */
 #include "paddle/pten/api/lib/kernel_dispatch.h"
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 
 PT_DECLARE_KERNEL(copy, CPU, ALL_LAYOUT);
 
diff --git a/paddle/pten/include/core.h b/paddle/pten/include/core.h
deleted file mode 100644
index 9a042753d1f738b362275f1df03bf3cac6b285a7..0000000000000000000000000000000000000000
--- a/paddle/pten/include/core.h
+++ /dev/null
@@ -1,22 +0,0 @@
-/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#pragma once
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/core/convert_utils.h"
-#include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/core/kernel_context.h"
-#include "paddle/pten/core/kernel_factory.h"
-#include "paddle/pten/core/tensor_meta.h"
diff --git a/paddle/pten/include/infermeta.h b/paddle/pten/include/infermeta.h
deleted file mode 100644
index 5e356dd37c03e2b3b67b3316e0df5c62d5a64e82..0000000000000000000000000000000000000000
--- a/paddle/pten/include/infermeta.h
+++ /dev/null
@@ -1,21 +0,0 @@
-/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#pragma once
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/infermeta/binary.h"
-#include "paddle/pten/infermeta/multiary.h"
-#include "paddle/pten/infermeta/nullary.h"
-#include "paddle/pten/infermeta/unary.h"
diff --git a/paddle/pten/include/math.h b/paddle/pten/include/math.h
deleted file mode 100644
index a4fb7f4d98fafb341f45dc420c74c40392046aa5..0000000000000000000000000000000000000000
--- a/paddle/pten/include/math.h
+++ /dev/null
@@ -1,39 +0,0 @@
-/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#pragma once
-
-// See Note: [ How do we organize the kernel directory ]
-#include "paddle/pten/api/lib/utils/storage.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/kernels/scale_kernel.h"
-
-namespace pten {
-
-template <typename T, typename ContextT>
-DenseTensor Scale(const ContextT& dev_ctx,
-                  const DenseTensor& x,
-                  const Scalar& scale,
-                  float bias,
-                  bool bias_after_scale) {
-  auto out_meta = UnchangedInferMeta(x.meta());
-  pten::DenseTensor dense_out(
-      pten::make_intrusive<paddle::experimental::SharedStorage>(
-          dev_ctx.GetPlace()),
-      std::move(out_meta));
-  Scale<T, ContextT>(dev_ctx, x, scale, bias, bias_after_scale, &dense_out);
-  return dense_out;
-}
-
-}  // namespace pten
diff --git a/paddle/pten/kernels/complex_kernel.h b/paddle/pten/kernels/complex_kernel.h
index 9dd3d457e4a2608b02396ed3326fb49ab0570eaa..b6074f117ea14ad6b970081f3cde1e21798f3b14 100644
--- a/paddle/pten/kernels/complex_kernel.h
+++ b/paddle/pten/kernels/complex_kernel.h
@@ -15,9 +15,6 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
-#include "paddle/pten/kernels/empty_kernel.h"
-
 #include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
diff --git a/paddle/pten/kernels/cpu/scale_kernel.cc b/paddle/pten/kernels/cpu/scale_kernel.cc
index fe9a0a033bced75e46be6a0de763d6d3e6ef86ec..0582fb87b4457dc6a50e815526b3e38b08ad5000 100644
--- a/paddle/pten/kernels/cpu/scale_kernel.cc
+++ b/paddle/pten/kernels/cpu/scale_kernel.cc
@@ -13,18 +13,48 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/pten/kernels/scale_kernel.h"
-#include "paddle/pten/kernels/impl/scale_kernel_impl.h"
 
 #include "paddle/pten/backends/cpu/cpu_context.h"
+#include "paddle/pten/common/scalar.h"
+#include "paddle/pten/core/dense_tensor.h"
 #include "paddle/pten/core/kernel_registry.h"
+#include "paddle/pten/kernels/funcs/eigen/common.h"
 
 // See Note [ Why still include the fluid headers? ]
+#include "paddle/fluid/operators/eigen/eigen_function.h"
 #include "paddle/fluid/platform/bfloat16.h"
+namespace pten {
+
+template <typename T, typename Context>
+void ScaleKernel(const Context& dev_ctx,
+                 const DenseTensor& x,
+                 const Scalar& scale,
+                 float bias,
+                 bool bias_after_scale,
+                 DenseTensor* out) {
+  // calc
+  out->mutable_data<T>();
+  auto eigen_out = pten::EigenVector<T>::Flatten(*out);
+  auto eigen_x = pten::EigenVector<T>::Flatten(x);
+  auto& dev = *dev_ctx.eigen_device();
+  // TODO(chenweihang): now the eigen function here need the dtype of scale,
+  // eigen_x, bias should be same, so here need cast for two scalar arg,
+  // maybe we declare that the type of scale and bias is T?
+  paddle::operators::EigenScale<std::decay_t<decltype(dev)>, T>::Eval(
+      dev,
+      eigen_out,
+      eigen_x,
+      scale.to<T>(),
+      static_cast<T>(bias),
+      bias_after_scale);
+}
+
+}  // namespace pten
 
 PT_REGISTER_CTX_KERNEL(scale,
                        CPU,
                        ALL_LAYOUT,
-                       pten::Scale,
+                       pten::ScaleKernel,
                        float,
                        double,
                        paddle::platform::bfloat16,
diff --git a/paddle/pten/kernels/flatten_kernel.h b/paddle/pten/kernels/flatten_kernel.h
index a67e66fac41300eca882f2deb7f71a14d946b839..c974fda1ed36322c021e5695e536277def5986c3 100644
--- a/paddle/pten/kernels/flatten_kernel.h
+++ b/paddle/pten/kernels/flatten_kernel.h
@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {
diff --git a/paddle/pten/kernels/gpu/scale_kernel.cu b/paddle/pten/kernels/gpu/scale_kernel.cu
index 68574c063e77f34988e468c5f46fcbfabba22b88..ff7e2a6ed284c1e2192a4c13f5ea59d737160a5d 100644
--- a/paddle/pten/kernels/gpu/scale_kernel.cu
+++ b/paddle/pten/kernels/gpu/scale_kernel.cu
@@ -44,12 +44,12 @@ struct ScaleFunctor {
 };
 
 template <typename T, typename ContextT>
-void Scale(const ContextT& dev_ctx,
-           const DenseTensor& x,
-           const Scalar& scale,
-           float bias,
-           bool bias_after_scale,
-           DenseTensor* out) {
+void ScaleKernel(const ContextT& dev_ctx,
+                 const DenseTensor& x,
+                 const Scalar& scale,
+                 float bias,
+                 bool bias_after_scale,
+                 DenseTensor* out) {
   std::vector<const DenseTensor*> inputs;
   std::vector<DenseTensor*> outputs;
   inputs.emplace_back(&x);
@@ -67,7 +67,7 @@ void Scale(const ContextT& dev_ctx,
 PT_REGISTER_CTX_KERNEL(scale,
                        GPU,
                        ALL_LAYOUT,
-                       pten::Scale,
+                       pten::ScaleKernel,
                        float,
                        double,
                        paddle::platform::float16,
diff --git a/paddle/pten/kernels/impl/matmul_grad_kernel_impl.h b/paddle/pten/kernels/impl/matmul_grad_kernel_impl.h
index 802cc019d78c5178c4c28b07586aa0ee8fbcd276..b1bae78ddc5fa27ab60d3065a9c2e1455113dfa0 100644
--- a/paddle/pten/kernels/impl/matmul_grad_kernel_impl.h
+++ b/paddle/pten/kernels/impl/matmul_grad_kernel_impl.h
@@ -14,8 +14,7 @@ limitations under the License. */
 
 #pragma once
 
-// #include "paddle/pten/kernels/complex_kernel.h"
-#include "paddle/pten/include/math.h"
+#include "paddle/pten/kernels/complex_kernel.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 #include "paddle/pten/kernels/impl/dot_grad_kernel_impl.h"
 #include "paddle/pten/kernels/impl/matmul_kernel_impl.h"
diff --git a/paddle/pten/kernels/impl/scale_kernel_impl.h b/paddle/pten/kernels/impl/scale_kernel_impl.h
deleted file mode 100644
index 2e0b158b36b8dd9ce10f447e54c9049f48919dba..0000000000000000000000000000000000000000
--- a/paddle/pten/kernels/impl/scale_kernel_impl.h
+++ /dev/null
@@ -1,50 +0,0 @@
-/* Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License. */
-
-#pragma once
-
-#include "paddle/pten/common/scalar.h"
-#include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/kernels/funcs/eigen/common.h"
-
-// See Note [ Why still include the fluid headers? ]
-#include "paddle/fluid/operators/eigen/eigen_function.h"
-
-namespace pten {
-
-template <typename T, typename Context>
-void Scale(const Context& dev_ctx,
-           const DenseTensor& x,
-           const Scalar& scale,
-           float bias,
-           bool bias_after_scale,
-           DenseTensor* out) {
-  // calc
-  out->mutable_data<T>();
-  auto eigen_out = pten::EigenVector<T>::Flatten(*out);
-  auto eigen_x = pten::EigenVector<T>::Flatten(x);
-  auto& dev = *dev_ctx.eigen_device();
-  // TODO(chenweihang): now the eigen function here need the dtype of scale,
-  // eigen_x, bias should be same, so here need cast for two scalar arg,
-  // maybe we declare that the type of scale and bias is T?
-  paddle::operators::EigenScale<std::decay_t<decltype(dev)>, T>::Eval(
-      dev,
-      eigen_out,
-      eigen_x,
-      scale.to<T>(),
-      static_cast<T>(bias),
-      bias_after_scale);
-}
-
-}  // namespace pten
diff --git a/paddle/pten/kernels/math_kernel.h b/paddle/pten/kernels/math_kernel.h
index f87d0a31b470bc0237e6b15fa1aa7eabab4bc6c7..e01103fc5b84757876b9a84e3794d47c51bcfcdd 100644
--- a/paddle/pten/kernels/math_kernel.h
+++ b/paddle/pten/kernels/math_kernel.h
@@ -16,7 +16,8 @@ limitations under the License. */
 
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/binary.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {
diff --git a/paddle/pten/kernels/reshape_kernel.h b/paddle/pten/kernels/reshape_kernel.h
index faa51c69ad17c282aac4187fc9860169685680c3..293f6cd2baf61dea1d9d40591690e2ac2ef76dac 100644
--- a/paddle/pten/kernels/reshape_kernel.h
+++ b/paddle/pten/kernels/reshape_kernel.h
@@ -16,7 +16,7 @@ limitations under the License. */
 
 #include "paddle/pten/common/scalar_array.h"
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {
diff --git a/paddle/pten/kernels/scale_kernel.h b/paddle/pten/kernels/scale_kernel.h
index 5908050029c7ae683b9f64074f9f54022bd994ce..ba16db566b8bb2a7f38ff2792287dac4e46cc08c 100644
--- a/paddle/pten/kernels/scale_kernel.h
+++ b/paddle/pten/kernels/scale_kernel.h
@@ -16,15 +16,29 @@ limitations under the License. */
 
 #include "paddle/pten/common/scalar.h"
 #include "paddle/pten/core/dense_tensor.h"
-
+#include "paddle/pten/infermeta/unary.h"
+#include "paddle/pten/kernels/empty_kernel.h"
 namespace pten {
 
 template <typename T, typename Context>
-void Scale(const Context& dev_ctx,
-           const DenseTensor& x,
-           const Scalar& scale,
-           float bias,
-           bool bias_after_scale,
-           DenseTensor* out);
+void ScaleKernel(const Context& dev_ctx,
+                 const DenseTensor& x,
+                 const Scalar& scale,
+                 float bias,
+                 bool bias_after_scale,
+                 DenseTensor* out);
+
+template <typename T, typename ContextT>
+DenseTensor Scale(const ContextT& dev_ctx,
+                  const DenseTensor& x,
+                  const Scalar& scale,
+                  float bias,
+                  bool bias_after_scale) {
+  auto out_meta = UnchangedInferMeta(x.meta());
+  auto dense_out = pten::Empty<T, ContextT>(dev_ctx, std::move(out_meta));
+  ScaleKernel<T, ContextT>(
+      dev_ctx, x, scale, bias, bias_after_scale, &dense_out);
+  return dense_out;
+}
 
 }  // namespace pten
diff --git a/paddle/pten/kernels/sign_kernel.h b/paddle/pten/kernels/sign_kernel.h
index ba205fc96a15caba739e41be7ef1ed1bc2f0c0a4..304b640d2af69f1f16dba6f6455656836d500516 100644
--- a/paddle/pten/kernels/sign_kernel.h
+++ b/paddle/pten/kernels/sign_kernel.h
@@ -15,7 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/pten/core/dense_tensor.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/empty_kernel.h"
 
 namespace pten {
diff --git a/paddle/pten/tests/api/scale_api.h b/paddle/pten/tests/api/scale_api.h
index d525b305c7409b76907ec2a70dcac940c9d6ccaa..41143826c45d8ba8bf3caa19c831a3604c776fa7 100644
--- a/paddle/pten/tests/api/scale_api.h
+++ b/paddle/pten/tests/api/scale_api.h
@@ -23,8 +23,7 @@
 #include "paddle/pten/common/scalar.h"
 #include "paddle/pten/common/scalar_array.h"
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/scale_kernel.h"
 
 namespace paddle {
@@ -92,42 +91,42 @@ static void ScaleCPU(DataType kernel_dtype,
                      pten::DenseTensor* dense_out) {
   switch (kernel_dtype) {
     case pten::DataType::FLOAT64: {
-      pten::Scale<double>(
+      pten::ScaleKernel<double>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::FLOAT32: {
-      pten::Scale<float>(
+      pten::ScaleKernel<float>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::BFLOAT16: {
-      pten::Scale<paddle::platform::bfloat16>(
+      pten::ScaleKernel<paddle::platform::bfloat16>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT64: {
-      pten::Scale<int64_t>(
+      pten::ScaleKernel<int64_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT32: {
-      pten::Scale<int32_t>(
+      pten::ScaleKernel<int32_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT16: {
-      pten::Scale<int16_t>(
+      pten::ScaleKernel<int16_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT8: {
-      pten::Scale<int8_t>(
+      pten::ScaleKernel<int8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::UINT8: {
-      pten::Scale<uint8_t>(
+      pten::ScaleKernel<uint8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
@@ -151,42 +150,42 @@ static void ScaleGPU(DataType kernel_dtype,
                      pten::DenseTensor* dense_out) {
   switch (kernel_dtype) {
     case pten::DataType::FLOAT64: {
-      pten::Scale<double>(
+      pten::ScaleKernel<double>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::FLOAT32: {
-      pten::Scale<float>(
+      pten::ScaleKernel<float>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::FLOAT16: {
-      pten::Scale<paddle::platform::float16>(
+      pten::ScaleKernel<paddle::platform::float16>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT64: {
-      pten::Scale<int64_t>(
+      pten::ScaleKernel<int64_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT32: {
-      pten::Scale<int32_t>(
+      pten::ScaleKernel<int32_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT16: {
-      pten::Scale<int16_t>(
+      pten::ScaleKernel<int16_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::INT8: {
-      pten::Scale<int8_t>(
+      pten::ScaleKernel<int8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
     case pten::DataType::UINT8: {
-      pten::Scale<uint8_t>(
+      pten::ScaleKernel<uint8_t>(
           dev_ctx, x, pten::Scalar(scale), bias, bias_after_scale, dense_out);
       break;
     }
diff --git a/paddle/pten/tests/kernels/test_scale_dev_api.cc b/paddle/pten/tests/kernels/test_scale_dev_api.cc
index ac2922b36f2059f81a1233e94d575dafa2011745..fe26f56552b05fc0d27d7af4e9de36a029f0a740 100644
--- a/paddle/pten/tests/kernels/test_scale_dev_api.cc
+++ b/paddle/pten/tests/kernels/test_scale_dev_api.cc
@@ -15,7 +15,7 @@ limitations under the License. */
 #include <gtest/gtest.h>
 #include <memory>
 
-#include "paddle/pten/include/math.h"
+#include "paddle/pten/kernels/scale_kernel.h"
 
 #include "paddle/pten/api/lib/utils/allocator.h"
 #include "paddle/pten/core/dense_tensor.h"
diff --git a/python/paddle/utils/code_gen/api_gen.py b/python/paddle/utils/code_gen/api_gen.py
index 35720ae32fe3891b6f9ac30face388fc1c053b4b..e8539b11d14550dfa76eb9affd7da2ce2b7aed4c 100644
--- a/python/paddle/utils/code_gen/api_gen.py
+++ b/python/paddle/utils/code_gen/api_gen.py
@@ -345,8 +345,10 @@ def source_include(header_file_path):
 #include "paddle/pten/api/lib/kernel_dispatch.h"
 #include "paddle/pten/api/lib/utils/storage.h"
 #include "paddle/pten/core/kernel_registry.h"
-#include "paddle/pten/include/core.h"
-#include "paddle/pten/include/infermeta.h"
+#include "paddle/pten/infermeta/binary.h"
+#include "paddle/pten/infermeta/multiary.h"
+#include "paddle/pten/infermeta/nullary.h"
+#include "paddle/pten/infermeta/unary.h"
 #include "paddle/pten/kernels/declarations.h"
 """