Not as "out of the box" as you seem to think. What you're describing is a sort of crude, clumsy, weak pre-encipherment. Instead of the crude, clumsy, Heath Robinson crap you suggest, why not just use a different modern cipher for your pre-encryption obfuscation? Why not pick something widely believed to be secure in its own right and designed to produce ciphertext which is demonstrably statistically/cryptographically indistinguishable from random data? Something which has withstood decades of analysis itself. Then encipher its statistically/cryptographically pseudorandom output using a different cipher and unrelated key. Just as Truecrypt used to do... Before it was harried into oblivion. There's been some interesting work done in proving the worth of such approaches and how they compound, useful in the light of offputting thought experiments like "evil ciphers" ...and which might go some way to explain what happened to TC.
Better than entrusting your "secret" in plaintext to the interwebs and Google Translate, wouldn't you agree?
Biting the hand that feeds IT
