Is it possible to remove the "batch" dimension of input after training?

Question

When I train the image classification model, the input shape is (None, Height, Width, Channels).

After train, I need to deploy it by tf-serving, and I still need to post a tensor with shape (None, Height, Width, Channels) to the production server.

Is it possible make it accepts (Height, Width, Channels) without reshape the single image input?