Then follow the leaders. Do it like big Softies do it. Make your code base huge, redundant with lots of legacy code and multiple dependencies that do not follow any logic. You must follow only bad coding practices at all cost. Buggy code helps a lot. Your code has to be so bad that you will not be able to understand what is going on after a while. This will definitely put a high price on any attempts to reverse engineer. Another way of doing it is to decouple the "view" from the "model" but to better secure it you would have to run secure server service yourself which is not easy.
I am not an expert in this but one thing I know is that there is no magic software that secures it. It is lots of IT work setting up properly OS and server then working on permissions, firewalls, testing for vulnerabilities, monitoring and so on. It is full time job and on top of that having fixed IP address costs more money. It truly requires certain level of expertise that not even average IT people have. In this scenario one probably has to settle on cloud based service keeping in mind that your code will be shared with NSA and others. The upside is that they will have backup copy of it just in case. Since they are able to get to any machine anyway doing it in the cloud lowers your profile since you demonstrate that you trust cloud services as opposed to someone paranoid running their own server.
Why not split the code up into different parts, and put each piece of code on a different machine on your local network? Each machine communicates with each other serially. I'm not an expert on networks and firewalls but you ensure that only one machine communicates with the internet and outside world, and this is the one that receives the incoming price data and sends out the buy and sell signals. It only computes the last piece of computation. The other machines are configured to only "talk" to each other, not the outside world, and only do the work necessary for their particular computation. So you have established this convoluted chain of machines (and quite deliberately so). Surely a network engineer could set this up for a few thousand$? You hire them for a week or so (and then you shoot them in the back as they leave your home if you're really paranoid ) They don't need to see the code, they just set up the network for you. They show you how each component in the chain communicates with each other. The code is not in one piece, so for all intents and purposes you have an impenetrable chain. Each machine only receives the data that it requires for the next calculation, which it then forwards on to the next machine in the chain. I'm not an IT expert so I might be talking out of my backside, but this seems to me a logical way to keep the code as safe as possible. As others have mentioned, you can convert some code into DLLs, use encryption at each step, use different operating systems on different machines, Linux etc and you basically put in as many obstacles as you can in to thwart any wannabee hacker. If they get past this convoluted chain of affairs, they probably deserve to see your code! There might be some flaws in the above, since as I said, I'm NOT an IT expert, but maybe it gives you some ideas. Maybe others can point out the flaws in my proposal? It all falls down of course if your system is latency dependent, but if you can wait a few seconds or so before firing off buy and sell signals, the additional delay caused by the daisy chain doesn't matter.