Compatibility Guarantees with tf-lite

We are trying to add support for libedgetpu in our software, and our options are:

1) Compile libedgetpu + tflite into a single static library (A bit of a pain with bazel currently) for our program and modify the interpreter with edgetpu delegate only when libedgetpu finds devices:

              context->edgetpu_devices = edgetpu_list_devices(&(context->edgetpu_device_count));

                if (context->edgetpu_device_count > 0) {
                    LOG_D("Edgetpu devices found. Trying to modify the interpreter to use the edgetpu delegate.");
                    const auto& device = context->edgetpu_devices[0];
                    TfLiteDelegate* delegate = edgetpu_create_delegate(device.type, device.path, nullptr, 0);
                    context->interpreter->ModifyGraphWithDelegate(std::unique_ptr<TfLiteDelegate, decltype(&edgetpu_free_delegate)>(delegate, &edgetpu_free_delegate));
                }

2) Compile tflite into our program, try to dlopen libedgetpu from the host system, and take the same code path. This option is attractive to us because we won't have to worry about statically linking / distributing libusb and licensing issues that come from that, build system becomes slightly easier.

We are wondering

If approach 2) is a feasible approach, and what versioning constraints about tflite vs. libedgetpu should we be wary about?
As per https://www.tensorflow.org/guide/versions , Can we assume that libedgetpu released along with a tflite version is compatible with all the versions of tflite in the same major version?
Is there a simpler solution other than the 2 above mentioned solutions?

google-coral / libedgetpu

Compatibility Guarantees with tf-lite #26