WRITING CUSTOM OP FOR EDGE_TPU INFERENCE?

Question

I am planning to add some custom operators to my quantized model on EDGE TPU. How can I include and self write that operator when running the EDGE TPU compiler? I have looke