diff --git a/paddle/fluid/lite/core/kernel.h b/paddle/fluid/lite/core/kernel.h
index c09158045c5520728cbea81df4c32c1435fa862d..bfabd87baff0a7584644322dd096c60b7b3e3a61 100644
--- a/paddle/fluid/lite/core/kernel.h
+++ b/paddle/fluid/lite/core/kernel.h
@@ -150,7 +150,7 @@ class KernelBase {
 // MKLDNN, plain CUDA C implementations.
 template <TargetType Target, PrecisionType Precision,
           DataLayoutType DataLayout = DataLayoutType::kNCHW>
-class OpKernel : public KernelBase {
+class KernelLite : public KernelBase {
  public:
   // Set runtime context.
   void SetContext(std::unique_ptr<KernelContext>&& ctx) { ctx_ = ctx; }
@@ -166,15 +166,15 @@ class OpKernel : public KernelBase {
 
   void Touch() {}
 
-  OpKernel() = default;
-  virtual ~OpKernel() = default;
+  KernelLite() = default;
+  virtual ~KernelLite() = default;
 
  protected:
   std::unique_ptr<KernelContext> ctx_;
 };
 
 template <TargetType Target, PrecisionType Precision, DataLayoutType DataLayout>
-std::string OpKernel<Target, Precision, DataLayout>::name() const {
+std::string KernelLite<Target, Precision, DataLayout>::name() const {
   return op_type() + ":" + TargetToStr(Target) + "/" +
          PrecisionToStr(Precision) + "/" + DataLayoutToStr(DataLayout);
 }
diff --git a/paddle/fluid/lite/core/kernel_test.cc b/paddle/fluid/lite/core/kernel_test.cc
index 045340d04a04e2fea3f6d67189df5d0d2f9661ff..c41b598a9fee93f13aa79479e108a2f5431fee62 100644
--- a/paddle/fluid/lite/core/kernel_test.cc
+++ b/paddle/fluid/lite/core/kernel_test.cc
@@ -21,7 +21,7 @@ namespace lite {
 namespace core {
 
 int test_code{-1};
-class SomeKernel : public OpKernel<TARGET(kHost), PRECISION(kFloat)> {
+class SomeKernel : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
  public:
   void Run() override {
     LOG(INFO) << "SomeKernel executed";
diff --git a/paddle/fluid/lite/core/op_registry.h b/paddle/fluid/lite/core/op_registry.h
index fb656c3a4ca3645aecbf2e64ba2ffa818c8b872a..6b8b0e038719cc2ea1fcc8a75278a91b14dd10c0 100644
--- a/paddle/fluid/lite/core/op_registry.h
+++ b/paddle/fluid/lite/core/op_registry.h
@@ -52,7 +52,7 @@ class OpLiteRegistor : public Registor<OpClass> {
 
 template <TargetType Target, PrecisionType Precision, DataLayoutType Layout>
 using KernelRegistryForTarget =
-    Factory<OpKernel<Target, Precision, Layout>, std::unique_ptr<KernelBase>>;
+    Factory<KernelLite<Target, Precision, Layout>, std::unique_ptr<KernelBase>>;
 
 class KernelRegistry final {
  public:
diff --git a/paddle/fluid/lite/kernels/cuda/io_copy_compute.cc b/paddle/fluid/lite/kernels/cuda/io_copy_compute.cc
index 01b62374df925be1d7317d347c998c03fbffe6f7..493c8a9181ef299e72b6e905dfb23c96c53b51cc 100644
--- a/paddle/fluid/lite/kernels/cuda/io_copy_compute.cc
+++ b/paddle/fluid/lite/kernels/cuda/io_copy_compute.cc
@@ -42,7 +42,7 @@ void CopyToHostSync(void* target, const void* source, size_t size) {
  * This kernel copies a tensor from host to CUDA space.
  */
 class IoCopyHostToCudaCompute
-    : public OpKernel<TARGET(kCUDA), PRECISION(kAny), DATALAYOUT(kAny)> {
+    : public KernelLite<TARGET(kCUDA), PRECISION(kAny), DATALAYOUT(kAny)> {
  public:
   void Run() override {
     auto& param = Param<operators::IoCopyParam>();
@@ -77,7 +77,7 @@ class IoCopyHostToCudaCompute
  * This kernel copies a tensor from CUDA to host space.
  */
 class IoCopyCudaToHostCompute
-    : public OpKernel<TARGET(kCUDA), PRECISION(kAny), DATALAYOUT(kAny)> {
+    : public KernelLite<TARGET(kCUDA), PRECISION(kAny), DATALAYOUT(kAny)> {
  public:
   void Run() override {
     auto& param = Param<operators::IoCopyParam>();
diff --git a/paddle/fluid/lite/kernels/cuda/mul_compute.h b/paddle/fluid/lite/kernels/cuda/mul_compute.h
index b7419e99ede808293d966ced6819c67688ad8cce..4b5998fd2fb9e9ea08de9951d78ebc676204a51b 100644
--- a/paddle/fluid/lite/kernels/cuda/mul_compute.h
+++ b/paddle/fluid/lite/kernels/cuda/mul_compute.h
@@ -30,7 +30,7 @@ void mul_compute(const lite::cuda::Blas<float>& blas, const T* x, int x_h,
              nullptr, out, x_h);
 }
 
-class MulCompute : public OpKernel<TARGET(kCUDA), PRECISION(kFloat)> {
+class MulCompute : public KernelLite<TARGET(kCUDA), PRECISION(kFloat)> {
  public:
   using param_t = operators::MulParam;
 
diff --git a/paddle/fluid/lite/kernels/host/fc_compute.h b/paddle/fluid/lite/kernels/host/fc_compute.h
index d835e96d4093a5173190e2e0929bb2180404d74c..78710c1202d58560ab04a8028b4e311e966b60ce 100644
--- a/paddle/fluid/lite/kernels/host/fc_compute.h
+++ b/paddle/fluid/lite/kernels/host/fc_compute.h
@@ -23,7 +23,7 @@ namespace lite {
 namespace kernels {
 namespace host {
 
-class FcCompute : public OpKernel<TARGET(kHost), PRECISION(kFloat)> {
+class FcCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
  public:
   using param_t = operators::FcParam;
 
diff --git a/paddle/fluid/lite/kernels/host/feed_compute.cc b/paddle/fluid/lite/kernels/host/feed_compute.cc
index b1cef73276b2da8287ff12c5bba4130478a2c004..0df9e8f429da20021dd9f450b20b07fe88171525 100644
--- a/paddle/fluid/lite/kernels/host/feed_compute.cc
+++ b/paddle/fluid/lite/kernels/host/feed_compute.cc
@@ -21,7 +21,7 @@ namespace kernels {
 namespace host {
 
 class FeedCompute
-    : public OpKernel<TARGET(kHost), PRECISION(kAny), DATALAYOUT(kAny)> {
+    : public KernelLite<TARGET(kHost), PRECISION(kAny), DATALAYOUT(kAny)> {
  public:
   using param_t = operators::FeedParam;
 
diff --git a/paddle/fluid/lite/kernels/host/fetch_compute.cc b/paddle/fluid/lite/kernels/host/fetch_compute.cc
index 4d640feff3744bf5eb8eb0447eb1b2ccd30cae9f..e23b540e6441efd14f082bf72a2d37982cdf6894 100644
--- a/paddle/fluid/lite/kernels/host/fetch_compute.cc
+++ b/paddle/fluid/lite/kernels/host/fetch_compute.cc
@@ -21,7 +21,7 @@ namespace kernels {
 namespace host {
 
 class FetchCompute
-    : public OpKernel<TARGET(kHost), PRECISION(kAny), DATALAYOUT(kAny)> {
+    : public KernelLite<TARGET(kHost), PRECISION(kAny), DATALAYOUT(kAny)> {
  public:
   using param_t = operators::FeedParam;
 
diff --git a/paddle/fluid/lite/kernels/host/mul_compute.cc b/paddle/fluid/lite/kernels/host/mul_compute.cc
index 2f6a95c83ba978da718ba58bd4c5e3f8dbc5f69a..9a61ffaa1c5463a8794ba5016ec019a3b4baea7f 100644
--- a/paddle/fluid/lite/kernels/host/mul_compute.cc
+++ b/paddle/fluid/lite/kernels/host/mul_compute.cc
@@ -35,7 +35,7 @@ void mul_compute_eigen(const T* x, int x_h, int x_w, const T* y, int y_h,
   Out = X * Y;
 }
 
-class MulCompute : public OpKernel<TARGET(kHost), PRECISION(kFloat)> {
+class MulCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
  public:
   using param_t = operators::MulParam;
 
diff --git a/paddle/fluid/lite/kernels/host/relu_compute.h b/paddle/fluid/lite/kernels/host/relu_compute.h
index 8afe1819184eb71a520ce1278bfe384752e42966..eec420e74bf2ce02fd9a201b03e2a3f771d70ff4 100644
--- a/paddle/fluid/lite/kernels/host/relu_compute.h
+++ b/paddle/fluid/lite/kernels/host/relu_compute.h
@@ -21,7 +21,7 @@ namespace lite {
 namespace kernels {
 namespace host {
 
-class ReluCompute : public OpKernel<TARGET(kHost), PRECISION(kFloat)> {
+class ReluCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
  public:
   void Run() override {
     auto& theparam = Param<operators::ReluParam>();
diff --git a/paddle/fluid/lite/kernels/host/scale_compute.cc b/paddle/fluid/lite/kernels/host/scale_compute.cc
index 4b260aecf87cbafbef32104bbecab7686c1b4642..b17498a3612b8217c5e65a758b3aef717559d1d7 100644
--- a/paddle/fluid/lite/kernels/host/scale_compute.cc
+++ b/paddle/fluid/lite/kernels/host/scale_compute.cc
@@ -31,7 +31,7 @@ void scale_compute(const T* x, T* out, int size, float scale, float bias,
   }
 }
 
-class ScaleCompute : public OpKernel<TARGET(kHost), PRECISION(kFloat)> {
+class ScaleCompute : public KernelLite<TARGET(kHost), PRECISION(kFloat)> {
  public:
   using param_t = operators::MulParam;