On a general basis, we don't want to encourage this kind of question:
References to code are not good descriptions of cryptographic algorithms. Good descriptions use the language of mathematics, not programming.
A description on a set of files on github is not permanent enough; if you ever change it, then this will make this question unreadable.
There is a very inefficient and tiresome cycle, which goes thus: "Hey, I don't know the subject, but I have this idea; is it secure ? -- No, it is not, for this reason. -- What if I add a +1 there ? -- No, does not help. -- Then what if I also put a +2 at that place ? -- ..." Experience shows that this kind of process never ends up with a good algorithm; however, it produces long-winded awful discussions. It is already bad enough on Usenet groups like
sci.crypt; on a Q&A site like this one, it will just not do.
From a cursory look at your files, though, it appears that you want to use Shamir's three-pass protocol. This requires a commutative cipher such that you can encrypt a message with key u, then with key v, and then decrypt with key u and v in that order, and get back the original message. Unfortunately, commutative ciphers are hard to make without a lot of mathematics, like modular exponentiations modulo big primes.
Your idea to do multiplications modulo 2n and to "encrypt" only message values x which are not invertible modulo 2n is not secure. Indeed, from the outside, the attacker sees x*u, x*v and x*u*v (all values modulo 2n). Since x is even, there are integers m and y such that y is odd and x = y*2m. u and v being odd (that's necessary for them to be invertible), the observed values are necessarily multiple of 2m and not 2m+1. In other words, expressed in binary they all end with exactly m zeros.
The attacker's goal is to recover x. He already knows m, as explained above. He wants y, which has length n-m bits. By simply dividing all the values he observed by 2m, he gets y*u, y*v and y*u*v, all values modulo 2n-m, and they are all odd, hence invertible. It then suffices to compute (y*u)*(y*v)/(y*u*v) (modulo 2n-m), which yields y. End game.
The above generalizes to all computations modulo some integer N when taking x to be non-invertible modulo N.
I encourage you to do your homework, i.e. read the Handbook of Applied Cryptography, which is a free download and very good reading (if a bit terse at times). Shamir's three-pass protocol is described in chapter 12, page 500. Mind the following passage:
While it might appear that any commutative cipher (i.e., cipher wherein the order of encryption and decryption is interchangeable)
would suffice in place of modular exponentiation in Protocol 12.22, caution is advised. For
example, use of the Vernam cipher (§1.5.4) would be totally insecure here, as the XOR of
the three exchanged messages would equal the key itself.
The worst sin in scientific research is to disregard previous research.