ecuda is a C++ wrapper around the CUDA C API designed to closely resemble and be functionally equivalent to the C++ Standard Template Library (STL) (algorithms, containers, and iterators). In addition to the familiar array and vector, matrix and cube containers are also included. The API is designed to work within kernel functions. CPU-only emulation is automatically supported when compiled with a C++ compiler other than nvcc (e.g. g++, clang). C++11 is utilized when enabled (but is not required). The API is header only, so it's portable and simple to install.
Version 2.0 has major improvements in all areas.
An overview of the project and the API documentation is available here.