Regarding choice of vision models, you might find this comparison helpful: https://github.com/explosion/prodigy-recipes/blob/master/image/tf_odapi/docs/model_performance.md . It was done as part of this object detection example:
I do think you should be able to find an image classification model that works acceptably on CPU. The benchmarks in the repo refer to object detection, so I think classification should be a bit more efficient.
Regarding the batch sizing and number of examples to update: ultimately you'll need to try this out on your problem and see what's working well.