Recently, I was reading: Are there any specific requirements for the function F in a Feistel cipher?, and the answer posted mentions a Feistel Cipher named Turtle, which uses a four-round Feistel scheme as a round function for a four-round Feistel scheme.
I wanted to know if doing so would provide better security than existing schemes, like Blowfish. And what if I use Blowfish Cipher itself as the round function (like in Turtle), will it, or, is it known to enhance security? If so, by how much?
I understood the idea of using cipher structures that are already known to be secure (as mentioned in the paper on Turtle that I linked above), but once the sub-Feistel Cipher is broken, the whole cipher should be broken, isn't it?
It's also the matter of the amount of processing required. It isn't heavy for general use, but it's a matter of concern when you expect your algorithm to be super fast.
I am currently working on an array of crypto algorithms for an inter-high-school science contest, which includes a Turtle like Feistel Cipher (Feistel Cipher using a sub-Feistel Cipher) and a fast PRNG that uses a Sponge Hash function that uses a Feistel Cipher as a function, and I thought I'd use the above-mentioned proposal of this cipher in the PRNG. However, I am a little skeptical about using it for the PRNG, as it might slow it down, and I am only trying to achieve an avalanche effect for the Hash function, which could also be accomplished by simpler Feistel construction. So, I also wanted to know if Turtle like algorithms is exceptionally good enough to be used in PRNGs, i.e., do they improve the unpredictability of the pseudo-random numbers generated.
P.S. I am insisting on using simpler constructions for the PRNG as I want it to be as fast as possible. I did a humiliatingly bad mistake in my last years' presentation for the same contest, where the then proposed PRNG took many seconds to generate a few numbers. This time I want it to be as fast as possible. So I will use the Turtle like proposal only if it increases the cryptographic security of the generator, or else I will resort to something simpler.