In the bazel, you can create a selectively built TensorFlow Lite based on your models. As described in the above guide document, the selective building will remove unused operators from the library and it will give you a slimmed version.
For example,
load(
"//tensorflow/lite:build_def.bzl",
"tflite_custom_cc_library",
)
# A selective built tflite for your models.
tflite_custom_cc_library(
name = "custom_tflite_lib",
models = [
"//path/.../model_a.tflite",
"//path/.../model_b.tflite",
],
)
You can also refer to this.