Yes, it's always good to leave X to somebody who knows how to do X. However, this ought not be left to "professional" cryptographers. After all, "professional"

Yes, it's always good to leave X to somebody who knows how to do X. However, this ought not be left to "professional" cryptographers. After all, "professional" merely means you're getting paid for it. History has shown again and again that many professional cryptographers do not know what their doing at all. From Crypto AG, over Wifi encryption standards, early SSL encryption, to cell phone encryption standards professional cryptography has failed again and again. There's probably more snake oil in commercial cryptography than in open-source hobbyist projects. Professional cryptographers, as well as intelligence agencies and police authorities around the world have a vested interest in perpetuating the myth that you shouldn't roll your own crypto, even though they know that e.g. certain combinations of cryptographic primitives or higher round numbers in existing algorithms are more secure at best and no less secure in the worst case. There are many other modifications that would potentially strengthen cryptography, e.g. triple encryption with minimum key or cascading ciphers. The advocacy against home-made cryptography started in response to the cyperhpunk movement in the 90s, which was perceived as a threat for a very simple reason: Nonstandard implementations need to be reverse-engineered individually, and this is very labour intensive and can become prohibitively expensive even for large well-funded government agencies. Use of standard libraries is therefore much more desirable, as it also makes it easier to tailor your side-channel attacks to the endpoint. Moreover, if there is a weakness in a library, it can be exploited on more systems. Also, you can attack the servers and machines of fewer developers, e.g. exchange binaries when they are downloaded.