I don't think this is possible. At least in TensorFlow 1.x, a lot of non-trivial parts (other than Operators and Gradients) are implemented in Python directly.

I don't think this is possible. At least in TensorFlow 1.x, a lot of non-trivial parts (other than Operators and Gradients) are implemented in Python directly. See their Java bindings, the functionalities exposed are miles far from the Python version.