This is more of a mental exercise for me than anything else. I've thought about doing something like this before, mostly to make a broken cipher a little more difficult to decrypt to plain text. I've simplified the strategy below, but in theory, you could use any length of replacement key first.
First, I would randomly choose a number of known length (say 4 digits). Let's use 1976 as an example. Then, I take the plain-text string, and do the following:
convert the string into a byte array. for the first byte, i add 1. for the second byte, i subtract 9. for the third byte, i add 7, and for the fourth byte, i subtract 6. I repeat this throughout the byte array, then rebuild the string from the byte array. Next, I concatenate the 1976 key to the front of the string. Once that is done, I use standard AES 256 encryption.
Now, in practice, I would use more than just adding and subtracting and a longer integer value, but that gives an example of the "randomization" of the string prior to encryption. The idea is to provide an extra layer of randomness to the encryption scheme. However, I'm not sure if this provides any real value as if the IV/Key has already been obtained then how hard is it really to figure out what I'm doing with the appended integer? Would it be better to use this post-encryption? I'm really curious what the community thinks of this idea. Does it have any merits? What are the drawbacks? How good/bad of an idea is it really?