diff --git a/CMakeLists.txt b/CMakeLists.txt
index c1b0061ce..cfc239394 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -61,7 +61,7 @@ if (MKL OR MKL_ROOT)
 endif()
 
 # There are about 30 files compiled for GPU. We merge them into less
-# for faster compilation. Set to 0 to mean no merging, or k to merge into k files. 
+# for faster compilation. Set to 0 to mean no merging, or k to merge into k files.
 if(NOT DEFINED GPU_NUMFILES)
   if(MSVC)
     set(GPU_NUMFILES 1)   # MSVC does serial compilation of CUDA, so minimize overhead
@@ -217,3 +217,8 @@ if(ENABLE_BOOST)
   add_subdirectory(tests)
   enable_testing()
 endif()
+
+export(TARGETS dynet NAMESPACE Dynet:: FILE DynetTargets.cmake)
+export(PACKAGE Dynet)
+file(WRITE ${CMAKE_BINARY_DIR}/DynetConfig.cmake
+    include(\"\${CMAKE_CURRENT_LIST_DIR}/DynetTargets.cmake\"))
diff --git a/dynet/CMakeLists.txt b/dynet/CMakeLists.txt
index cb3fab7e2..e21ecb25c 100644
--- a/dynet/CMakeLists.txt
+++ b/dynet/CMakeLists.txt
@@ -163,7 +163,7 @@ weight-decay.h
 if(ENABLE_BOOST)
   list(APPEND dynet_library_HDRS mp.h)
 endif()
-  
+
 set(dynet_gpu_mergeable_SRCS
     nodes-activations
     nodes-affinetransform
@@ -302,6 +302,9 @@ else()
   target_link_libraries(dynet ${LIBS})
 endif(WITH_CUDA_BACKEND)
 
+set_property(TARGET dynet APPEND PROPERTY INTERFACE_INCLUDE_DIRECTORIES
+    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/../>)
+
 install(FILES ${dynet_library_HDRS} DESTINATION include/dynet)
 install(TARGETS dynet DESTINATION lib)
 if(WITH_CUDA_BACKEND)
@@ -309,3 +312,4 @@ if(WITH_CUDA_BACKEND)
 endif(WITH_CUDA_BACKEND)
 
 # target_compile_features(dynet PRIVATE cxx_range_for)
+