I can see that for components heavily interfacing with high surface area things like encryption, hardware interfacing etc., but why would that be true for a relatively “pure” computational problem like an image codec? Bytes in, bytes out.