Smart pointers in __device__ and __global__ functions

When instantiating a new object inside a CUDA kernel, is it possible to use smart pointers (ie. make_shared<>())?

AFAIK smart pointers in C++ are part of std::memory

The base CUDA C++ language does not provide standard library support.

There is a NVIDIA-sponsored development to provide something like standard library support, but at the moment I don’t see <memory> or smart pointers there. (It’s possible I may have missed it)

A google search on “cuda smart pointer” yields some 3rd parties that have created their own implementations (that is not the only one).