Re-implementing crypto functions is generally a terrible idea, unless you really know what you are doing (and most likely you don't). There are so many differen

Re-implementing crypto functions is generally a terrible idea, unless you really know what you are doing (and most likely you don't). There are so many different relevant factors to consider, that single individuals or organizations are really unlikely to be able to cover reasonably close to completely. That's why standards and "well known" algorithms are a good choice. Also, "security through obscurity" is well known as a worst practice for crypto (probability that the algorithm will be reverse engineered or disclosed is all but negligible), so definitely not a good reason to implement any new algorithm.