Why not? The NDK/JNI calls are still in user space themselves. So what delineation are you trying to make here?

How userspace applications talk to the kernel subsystems, in a legal way without hacking the operating system architecture.