Well, the inevitable fate of cryptogams modules is to be integrated into some other framework. This implies that it's not given that performance is independent of context. In other words benchmarking is better done in target framework context, and it's the one that actually matters. I customarily provide 'cycles per processed byte' metrics in each module, and the real question should be if target framework achieves this mark. But this question can't be answered in advance. You have to run your own benchmarks... And if they fall short, tell me about it. [And if they overshoot, you're likely to fail to account for dynamic frequency scaling.]