diff --git a/lite/example/cpp_example/CMakeLists.txt b/lite/example/cpp_example/CMakeLists.txt
index 9012fded9a2d8bef2b43b0feb07c8d5203628708..f7227b62f1b491255b5dbd679a52c4ab1c42df15 100644
--- a/lite/example/cpp_example/CMakeLists.txt
+++ b/lite/example/cpp_example/CMakeLists.txt
@@ -1,5 +1,6 @@
 file (GLOB_RECURSE SOURCES ./*.cpp)
 add_executable(lite_examples  ${SOURCES})
+target_include_directories(lite_examples PUBLIC ./)
 
 if(LITE_BUILD_WITH_RKNPU)
     #rknn sdk1.0.0 depend on libc++_shared, use gold to remove NEEDED so symbol check
@@ -33,6 +34,7 @@ if(LITE_BUILD_WITH_RKNPU)
 endif()
 
 target_link_libraries(lite_examples_depends_shared lite_shared)
+target_include_directories(lite_examples_depends_shared PUBLIC ./)
 
 if(UNIX)
     if(APPLE OR ANDROID)
diff --git a/lite/example/cpp_example/example.h b/lite/example/cpp_example/example.h
index 0e5da8f5b386396b5f3c9d2174c3a79d94f3df63..8e07dcb105d89b73e2017513b1d4a41f4546cb02 100644
--- a/lite/example/cpp_example/example.h
+++ b/lite/example/cpp_example/example.h
@@ -49,57 +49,20 @@ ExampleFuncMap* get_example_function_map();
 
 bool register_example(std::string example_name, const ExampleFunc& fuction);
 
-template <int>
-struct Register;
-
-#if LITE_BUILD_WITH_MGE
-
-bool basic_load_from_path(const Args& args);
-bool basic_load_from_path_with_loader(const Args& args);
-bool basic_load_from_memory(const Args& args);
-bool cpu_affinity(const Args& args);
-bool network_share_same_weights(const Args& args);
-bool reset_input(const Args& args);
-bool reset_input_output(const Args& args);
-bool config_user_allocator(const Args& args);
-bool register_cryption_method(const Args& args);
-bool update_cryption_key(const Args& args);
-bool async_forward(const Args& args);
-bool set_input_callback(const Args& arg);
-bool set_output_callback(const Args& arg);
-
-bool picture_classification(const Args& arg);
-bool detect_yolox(const Args& arg);
-
-#if LITE_WITH_CUDA
-bool load_from_path_run_cuda(const Args& args);
-bool device_input(const Args& args);
-bool device_input_output(const Args& args);
-bool pinned_host_input(const Args& args);
-#endif
-#endif
-
 }  // namespace example
 }  // namespace lite
 
-#if LITE_BUILD_WITH_MGE
-bool basic_c_interface(const lite::example::Args& args);
-bool device_io_c_interface(const lite::example::Args& args);
-bool async_c_interface(const lite::example::Args& args);
-#endif
-
 #define CONCAT_IMPL(a, b)  a##b
 #define MACRO_CONCAT(a, b) CONCAT_IMPL(a, b)
 
 #define REGIST_EXAMPLE(name_, func_) REGIST_EXAMPLE_WITH_NUM(__COUNTER__, name_, func_)
 
-#define REGIST_EXAMPLE_WITH_NUM(number_, name_, func_)          \
-    template <>                                                 \
-    struct Register<number_> {                                  \
-        Register() { register_example(name_, func_); }          \
-    };                                                          \
-    namespace {                                                 \
-    Register<number_> MACRO_CONCAT(example_function_, number_); \
+#define REGIST_EXAMPLE_WITH_NUM(number_, name_, func_)                        \
+    struct Register_##func_ {                                                 \
+        Register_##func_() { lite::example::register_example(name_, func_); } \
+    };                                                                        \
+    namespace {                                                               \
+    Register_##func_ MACRO_CONCAT(func_, number_);                            \
     }
 
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/main.cpp b/lite/example/cpp_example/main.cpp
index b479b63d2b6cf3ea6038524090bcdbb96c683343..ff566ccb8d5e239e49caa448cccb94c16f37bbf6 100644
--- a/lite/example/cpp_example/main.cpp
+++ b/lite/example/cpp_example/main.cpp
@@ -60,7 +60,8 @@ bool lite::example::register_example(
         std::string example_name, const ExampleFunc& fuction) {
     auto map = get_example_function_map();
     if (map->find(example_name) != map->end()) {
-        printf("Error!!! This example is registed yet\n");
+        printf("example_name: %s Error!!! This example is registed yet\n",
+               example_name.c_str());
         return false;
     }
     (*map)[example_name] = fuction;
@@ -142,41 +143,5 @@ int main(int argc, char** argv) {
         return -1;
     }
 }
-namespace lite {
-namespace example {
-
-#if LITE_BUILD_WITH_MGE
-#if LITE_WITH_CUDA
-REGIST_EXAMPLE("load_from_path_run_cuda", load_from_path_run_cuda);
-#endif
-REGIST_EXAMPLE("basic_load_from_path", basic_load_from_path);
-REGIST_EXAMPLE("basic_load_from_path_with_loader", basic_load_from_path_with_loader);
-REGIST_EXAMPLE("basic_load_from_memory", basic_load_from_memory);
-REGIST_EXAMPLE("cpu_affinity", cpu_affinity);
-REGIST_EXAMPLE("register_cryption_method", register_cryption_method);
-REGIST_EXAMPLE("update_cryption_key", update_cryption_key);
-REGIST_EXAMPLE("network_share_same_weights", network_share_same_weights);
-REGIST_EXAMPLE("reset_input", reset_input);
-REGIST_EXAMPLE("reset_input_output", reset_input_output);
-REGIST_EXAMPLE("config_user_allocator", config_user_allocator);
-REGIST_EXAMPLE("async_forward", async_forward);
-REGIST_EXAMPLE("set_input_callback", set_input_callback);
-REGIST_EXAMPLE("set_output_callback", set_output_callback);
-
-REGIST_EXAMPLE("basic_c_interface", basic_c_interface);
-REGIST_EXAMPLE("device_io_c_interface", device_io_c_interface);
-REGIST_EXAMPLE("async_c_interface", async_c_interface);
-
-REGIST_EXAMPLE("picture_classification", picture_classification);
-REGIST_EXAMPLE("detect_yolox", detect_yolox);
-
-#if LITE_WITH_CUDA
-REGIST_EXAMPLE("device_input", device_input);
-REGIST_EXAMPLE("device_input_output", device_input_output);
-REGIST_EXAMPLE("pinned_host_input", pinned_host_input);
-#endif
-#endif
-}  // namespace example
-}  // namespace lite
 
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/basic.cpp b/lite/example/cpp_example/mge/basic.cpp
index 55e20270f9c0cf03b6519d71693025cf5b1c2686..069ddb93e6b9425e3bb8d6231c2416a1b39a1158 100644
--- a/lite/example/cpp_example/mge/basic.cpp
+++ b/lite/example/cpp_example/mge/basic.cpp
@@ -10,7 +10,7 @@
  */
 
 #include <thread>
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 #include <cstdio>
 
@@ -77,61 +77,8 @@ void output_data_info(std::shared_ptr<Network> network, size_t output_size) {
 }
 }  // namespace
 
-#if LITE_WITH_CUDA
-bool lite::example::load_from_path_run_cuda(const Args& args) {
-    std::string network_path = args.model_path;
-    std::string input_path = args.input_path;
-    set_log_level(LiteLogLevel::DEBUG);
-    //! config the network running in CUDA device
-    lite::Config config{false, -1, LiteDeviceType::LITE_CUDA};
-    //! set NetworkIO
-    NetworkIO network_io;
-    std::string input_name = "img0_comp_fullface";
-    bool is_host = false;
-    IO device_input{input_name, is_host};
-    network_io.inputs.push_back(device_input);
-    //! create and load the network
-    std::shared_ptr<Network> network = std::make_shared<Network>(config, network_io);
-    network->load_model(network_path);
-
-    std::shared_ptr<Tensor> input_tensor = network->get_input_tensor(0);
-    Layout input_layout = input_tensor->get_layout();
-
-    //! read data from numpy data file
-    auto src_tensor = parse_npy(input_path);
-
-    //! malloc the device memory
-    auto tensor_device = Tensor(LiteDeviceType::LITE_CUDA, input_layout);
-
-    //! copy to the device memory
-    tensor_device.copy_from(*src_tensor);
-
-    //! Now the device memory if filled with user input data, set it to the
-    //! input tensor
-    input_tensor->reset(tensor_device.get_memory_ptr(), input_layout);
-
-    //! forward
-    {
-        lite::Timer ltimer("warmup");
-        network->forward();
-        network->wait();
-        ltimer.print_used_time(0);
-    }
-    lite::Timer ltimer("forward_iter");
-    for (int i = 0; i < 10; i++) {
-        ltimer.reset_start();
-        network->forward();
-        network->wait();
-        ltimer.print_used_time(i);
-    }
-    //! get the output data or read tensor set in network_in
-    size_t output_size = network->get_all_output_name().size();
-    output_info(network, output_size);
-    output_data_info(network, output_size);
-    return true;
-}
-#endif
-bool lite::example::basic_load_from_path(const Args& args) {
+namespace {
+bool basic_load_from_path(const Args& args) {
     set_log_level(LiteLogLevel::DEBUG);
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
@@ -193,7 +140,7 @@ bool lite::example::basic_load_from_path(const Args& args) {
     return true;
 }
 
-bool lite::example::basic_load_from_path_with_loader(const Args& args) {
+bool basic_load_from_path_with_loader(const Args& args) {
     set_log_level(LiteLogLevel::DEBUG);
     lite::set_loader_lib_path(args.loader_path);
     std::string network_path = args.model_path;
@@ -251,7 +198,7 @@ bool lite::example::basic_load_from_path_with_loader(const Args& args) {
     return true;
 }
 
-bool lite::example::basic_load_from_memory(const Args& args) {
+bool basic_load_from_memory(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -307,7 +254,7 @@ bool lite::example::basic_load_from_memory(const Args& args) {
     return true;
 }
 
-bool lite::example::async_forward(const Args& args) {
+bool async_forward(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
     Config config;
@@ -366,7 +313,7 @@ bool lite::example::async_forward(const Args& args) {
     return true;
 }
 
-bool lite::example::set_input_callback(const Args& args) {
+bool set_input_callback(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
     Config config;
@@ -433,7 +380,7 @@ bool lite::example::set_input_callback(const Args& args) {
     return true;
 }
 
-bool lite::example::set_output_callback(const Args& args) {
+bool set_output_callback(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
     Config config;
@@ -500,7 +447,73 @@ bool lite::example::set_output_callback(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("basic_load_from_path", basic_load_from_path);
+REGIST_EXAMPLE("basic_load_from_path_with_loader", basic_load_from_path_with_loader);
+REGIST_EXAMPLE("basic_load_from_memory", basic_load_from_memory);
+REGIST_EXAMPLE("async_forward", async_forward);
+REGIST_EXAMPLE("set_input_callback", set_input_callback);
+REGIST_EXAMPLE("set_output_callback", set_output_callback);
+
+#if LITE_WITH_CUDA
+namespace {
+bool load_from_path_run_cuda(const Args& args) {
+    std::string network_path = args.model_path;
+    std::string input_path = args.input_path;
+    set_log_level(LiteLogLevel::DEBUG);
+    //! config the network running in CUDA device
+    lite::Config config{false, -1, LiteDeviceType::LITE_CUDA};
+    //! set NetworkIO
+    NetworkIO network_io;
+    std::string input_name = "img0_comp_fullface";
+    bool is_host = false;
+    IO device_input{input_name, is_host};
+    network_io.inputs.push_back(device_input);
+    //! create and load the network
+    std::shared_ptr<Network> network = std::make_shared<Network>(config, network_io);
+    network->load_model(network_path);
+
+    std::shared_ptr<Tensor> input_tensor = network->get_input_tensor(0);
+    Layout input_layout = input_tensor->get_layout();
+
+    //! read data from numpy data file
+    auto src_tensor = parse_npy(input_path);
+
+    //! malloc the device memory
+    auto tensor_device = Tensor(LiteDeviceType::LITE_CUDA, input_layout);
+
+    //! copy to the device memory
+    tensor_device.copy_from(*src_tensor);
+
+    //! Now the device memory if filled with user input data, set it to the
+    //! input tensor
+    input_tensor->reset(tensor_device.get_memory_ptr(), input_layout);
+
+    //! forward
+    {
+        lite::Timer ltimer("warmup");
+        network->forward();
+        network->wait();
+        ltimer.print_used_time(0);
+    }
+    lite::Timer ltimer("forward_iter");
+    for (int i = 0; i < 10; i++) {
+        ltimer.reset_start();
+        network->forward();
+        network->wait();
+        ltimer.print_used_time(i);
+    }
+    //! get the output data or read tensor set in network_in
+    size_t output_size = network->get_all_output_name().size();
+    output_info(network, output_size);
+    output_data_info(network, output_size);
+    return true;
+}
+}  // namespace
 
+REGIST_EXAMPLE("load_from_path_run_cuda", load_from_path_run_cuda);
+#endif
 #endif
 
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/cpu_affinity.cpp b/lite/example/cpp_example/mge/cpu_affinity.cpp
index ea936bf7a371a4ccb83b054b9a67e9c0502c228f..6d47c94929c879b3022f9727da2be070cb7ad86d 100644
--- a/lite/example/cpp_example/mge/cpu_affinity.cpp
+++ b/lite/example/cpp_example/mge/cpu_affinity.cpp
@@ -9,13 +9,14 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 
 using namespace lite;
 using namespace example;
 
-bool lite::example::cpu_affinity(const Args& args) {
+namespace {
+bool cpu_affinity(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -65,6 +66,9 @@ bool lite::example::cpu_affinity(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("cpu_affinity", cpu_affinity);
 #endif
 
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/cv/detect_yolox.cpp b/lite/example/cpp_example/mge/cv/detect_yolox.cpp
index b42485c08e81a49708fd802a3bb18df36a96b46d..e5f7755331b36174d422df58f39da7b1f8b9598b 100644
--- a/lite/example/cpp_example/mge/cv/detect_yolox.cpp
+++ b/lite/example/cpp_example/mge/cv/detect_yolox.cpp
@@ -10,7 +10,7 @@
  */
 
 #include <thread>
-#include "../../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 #include <cstdio>
 
@@ -289,6 +289,10 @@ void decode_outputs(
 void draw_objects(
         uint8_t* image, int width, int height, int channel,
         const std::vector<Object>& objects) {
+    (void)image;
+    (void)width;
+    (void)height;
+    (void)channel;
     for (size_t i = 0; i < objects.size(); i++) {
         const Object& obj = objects[i];
 
@@ -297,9 +301,7 @@ void draw_objects(
     }
 }
 
-}  // namespace
-
-bool lite::example::detect_yolox(const Args& args) {
+bool detect_yolox(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -332,6 +334,9 @@ bool lite::example::detect_yolox(const Args& args) {
     stbi_image_free(image);
     return 0;
 }
+}  // namespace
+
+REGIST_EXAMPLE("detect_yolox", detect_yolox);
 
 #endif
 
diff --git a/lite/example/cpp_example/mge/cv/picture_classification.cpp b/lite/example/cpp_example/mge/cv/picture_classification.cpp
index 84387e037dd350bb8348129173269aa5255b696b..970dc080f98edcf44e8af63da41a11db9c1cd401 100644
--- a/lite/example/cpp_example/mge/cv/picture_classification.cpp
+++ b/lite/example/cpp_example/mge/cv/picture_classification.cpp
@@ -10,7 +10,7 @@
  */
 
 #include <thread>
-#include "../../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 #include <cstdio>
 
@@ -80,9 +80,8 @@ void classfication_process(
     }
     printf("output tensor sum is %f\n", sum);
 }
-}  // namespace
 
-bool lite::example::picture_classification(const Args& args) {
+bool picture_classification(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -109,6 +108,9 @@ bool lite::example::picture_classification(const Args& args) {
            class_id, score);
     return 0;
 }
+}  // namespace
+
+REGIST_EXAMPLE("picture_classification", picture_classification);
 
 #endif
 
diff --git a/lite/example/cpp_example/mge/device_io.cpp b/lite/example/cpp_example/mge/device_io.cpp
index e7689ef5a194d08d991792e8cc642f0258cf8546..fd5626fb0a650a5f67f5d2a38e4888a4d13a3388 100644
--- a/lite/example/cpp_example/mge/device_io.cpp
+++ b/lite/example/cpp_example/mge/device_io.cpp
@@ -10,15 +10,17 @@
  */
 
 #include <thread>
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
+#include "misc.h"
 
 using namespace lite;
 using namespace example;
 
 #if LITE_WITH_CUDA
 
-bool lite::example::device_input(const Args& args) {
+namespace {
+bool device_input(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -73,7 +75,7 @@ bool lite::example::device_input(const Args& args) {
     return true;
 }
 
-bool lite::example::device_input_output(const Args& args) {
+bool device_input_output(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -136,7 +138,7 @@ bool lite::example::device_input_output(const Args& args) {
     return true;
 }
 
-bool lite::example::pinned_host_input(const Args& args) {
+bool pinned_host_input(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -181,6 +183,11 @@ bool lite::example::pinned_host_input(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("device_input", device_input);
+REGIST_EXAMPLE("device_input_output", device_input_output);
+REGIST_EXAMPLE("pinned_host_input", pinned_host_input);
 
 #endif
 #endif
diff --git a/lite/example/cpp_example/mge/lite_c_interface.cpp b/lite/example/cpp_example/mge/lite_c_interface.cpp
index 0424631ce0f8cd127d9bba7ea6ee16d02378ba88..e27e8a59f14d3f09b16dc4568a5f21575eca69be 100644
--- a/lite/example/cpp_example/mge/lite_c_interface.cpp
+++ b/lite/example/cpp_example/mge/lite_c_interface.cpp
@@ -9,7 +9,7 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 
-#include "../example.h"
+#include "example.h"
 #include "misc.h"
 #if LITE_BUILD_WITH_MGE
 #include "lite-c/global_c.h"
@@ -218,5 +218,10 @@ bool async_c_interface(const lite::example::Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+
+REGIST_EXAMPLE("basic_c_interface", basic_c_interface);
+REGIST_EXAMPLE("device_io_c_interface", device_io_c_interface);
+REGIST_EXAMPLE("async_c_interface", async_c_interface);
+
 #endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/network_share_weights.cpp b/lite/example/cpp_example/mge/network_share_weights.cpp
index 0a117d583f0fa0bc5e1303d5753f83067ca0646b..e206d394c4a5a68bc79a7dc1962251952698814b 100644
--- a/lite/example/cpp_example/mge/network_share_weights.cpp
+++ b/lite/example/cpp_example/mge/network_share_weights.cpp
@@ -9,13 +9,15 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 
 using namespace lite;
 using namespace example;
 
-bool lite::example::network_share_same_weights(const Args& args) {
+namespace {
+
+bool network_share_same_weights(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -75,5 +77,9 @@ bool lite::example::network_share_same_weights(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("network_share_same_weights", network_share_same_weights);
+
 #endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/reset_io.cpp b/lite/example/cpp_example/mge/reset_io.cpp
index cda27726a76d8b1ea52adbaab41e124525f96ee9..b6848195de0e01a7559d7368bc3526a924072312 100644
--- a/lite/example/cpp_example/mge/reset_io.cpp
+++ b/lite/example/cpp_example/mge/reset_io.cpp
@@ -9,13 +9,15 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 
 using namespace lite;
 using namespace example;
 
-bool lite::example::reset_input(const Args& args) {
+namespace {
+
+bool reset_input(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
     lite::Config config;
@@ -53,7 +55,7 @@ bool lite::example::reset_input(const Args& args) {
     return true;
 }
 
-bool lite::example::reset_input_output(const Args& args) {
+bool reset_input_output(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
     lite::Config config;
@@ -92,5 +94,10 @@ bool lite::example::reset_input_output(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("reset_input", reset_input);
+REGIST_EXAMPLE("reset_input_output", reset_input_output);
+
 #endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/user_allocator.cpp b/lite/example/cpp_example/mge/user_allocator.cpp
index 73b9b41df1756bf640d9cd1b94925009fa162aec..7f54743583fe492bbc6a29b171d6a4ce7c736caa 100644
--- a/lite/example/cpp_example/mge/user_allocator.cpp
+++ b/lite/example/cpp_example/mge/user_allocator.cpp
@@ -9,7 +9,7 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 using namespace lite;
 using namespace example;
@@ -42,9 +42,8 @@ public:
 #endif
     };
 };
-}  // namespace
 
-bool lite::example::config_user_allocator(const Args& args) {
+bool config_user_allocator(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -87,5 +86,9 @@ bool lite::example::config_user_allocator(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("config_user_allocator", config_user_allocator);
+
 #endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/lite/example/cpp_example/mge/user_cryption.cpp b/lite/example/cpp_example/mge/user_cryption.cpp
index 4a8579375f3000e1a2b7ec9488502352f3819bc9..02c8a7aceed226d547e7bc186bfcce0005bd6774 100644
--- a/lite/example/cpp_example/mge/user_cryption.cpp
+++ b/lite/example/cpp_example/mge/user_cryption.cpp
@@ -9,7 +9,7 @@
  * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  */
 
-#include "../example.h"
+#include "example.h"
 #if LITE_BUILD_WITH_MGE
 
 using namespace lite;
@@ -31,9 +31,8 @@ std::vector<uint8_t> decrypt_model(
         return {};
     }
 }
-}  // namespace
 
-bool lite::example::register_cryption_method(const Args& args) {
+bool register_cryption_method(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -75,7 +74,7 @@ bool lite::example::register_cryption_method(const Args& args) {
     return true;
 }
 
-bool lite::example::update_cryption_key(const Args& args) {
+bool update_cryption_key(const Args& args) {
     std::string network_path = args.model_path;
     std::string input_path = args.input_path;
 
@@ -120,5 +119,9 @@ bool lite::example::update_cryption_key(const Args& args) {
     printf("max=%e, sum=%e\n", max, sum);
     return true;
 }
+}  // namespace
+
+REGIST_EXAMPLE("register_cryption_method", register_cryption_method);
+REGIST_EXAMPLE("update_cryption_key", update_cryption_key);
 #endif
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}