AIM - To load a custom model which applies a NN and tf-df GBDT model one after the other

While trying to save and load a custom model which is a stacked version of an NN model and a GBDT model, while trying to register the model on inference sidecar I get the error ""Status: fail: NOT_FOUND: Op type not registered 'SimpleMLCreateModelResource' in binary running on 3bdaf418bc27""
When a YDF model is saved, some YDF-specific operations are written in the model graph, one of them is the SimpleMLCreateModelResource we've been getting in the error trace. The problem is that the tensorflow runtime in the inference sidecar does not contain the kernels for those operations; it only has a stripped down subset of vanilla TF operations.

One solution is to include the right inference kernel and ops from YDF as dependencies in the inference service's bazel files and rebuild the inference sidecar. The ops are registered statically when the TF binary runs, so there's no need to change the source code of the inference sidecar as long as the dependencies are working. tf-df is supported Natively in tf-serving 2.11+. Would using these versions of tf-serving solve the issue ?

Are there any alternatives to this?