The thing that intimidates me about games like GURPS or HERO is that they seem like buying a house just to have somewhere to place a chair or ordering a complete seven-course-meal just to get the dessert. Their premise is that they can do anything, but when you sit down to play a game you're usually just about something. It's one of the reasons why I'm more on the side of smaller, more focused games.
But what if there isn't a small, focused game for that something you want to do? Well, you can probably hack another game, or write one from scratch.
You can start by asking what the characters will be doing in this game. If you can't figure that out, then you have a problem. For example, for a Wild West game I might write down these activities:
Riding, Gambling, Shooting, Scouting, Trailblazing, Throwing, Handling Animals, Hunting, Fixing, Healing, Displaying Book Learning, Brawling, Sweet-talking
And that's the skill list right there. It's what I expect the characters to do, and I've got it covered. I might add one or two but I usually find that they fold right back into some existing skill and we don't want skills that are too situational (and useless) or too many skills in general. Another good thing about this is that it communicates clearly to the players what kind of activities their characters are supposed to be getting into.
You can add a generic or more specific set of abilities to that (the classic physical/mental/social trio that SO MANY games use for example), but these days I prefer a single layer of character competence instead of the rather common skill+stat combo.
The next step is resolution. You've probably played enough games from Risk to Poker to D&D that you have a relatively broad knowledge of how things can be resolved in games, with dice or cards or some other method. (If not, play more games!) It's good to have an understanding of the probabilities of the mechanic you're going to pick, as well as the various pros and cons. If we stick to dice (the most common in RPGs): for example dice pools are good because it's tactile (you feel in your hand how good your character is) and you can put extra mechanics on top where you can spend the successes as currency for extra effects or keep them for latter or whatever. Throwing more dice and summing them up is good because it creates a bellcurve, if you want that. Roll under percentile is good because it creates and immediate understanding of your probabilities for success. Etc. Of course all methods also have their drawbacks, depending on what you want.
One warning: don't get stuck on the mechanics of your game. I know very well from my own experience, from design forums and confessions of professionals that giving too much attention to how the dice work or starting from the math will kill your game. Deciding whether your game should use a d8 or a d12 for something is the least important part of design. Ok, not really, but it's needs to be said that way, to break you out of the loop.
Let's say I pick the percentile resolution method for my theoretical Western game. I just need to decide how many points the PCs can be built on, you give your character a name and we can play. If you have any experience running RPGs you can make it work no problem. That's a complete, playable game right there. However it's also a pretty poor game in the way a soup with only one kind of vegetable and no salt is poor. Not very flavorful and not very sustainable.
And that's where the real hard work of design begins (and often ends).