Is it possible to launch kernels on multiple devices with different back-ends? For instance, reading images from disk using a CPU and running computations on these images using a GPU with CUDA back-end.