Only the CPU, GPU and RAM can be overclocked.
One way I like to go when overclocking GPUs is to first find the limits of the GPU itself without touching the voltages. This means making slight 5-10 MHz bumps until artifacts begin to show, or the benchmark/game I'm running glitches somehow. You should first bump either the core or the memory, never both at the same time. Once you find your limits, back away 5-10 MHz from that and you should be fine.
To test stability, you should run a solid benchmark such as 3DMark or Valley. You could also run GPU-intensive games, such as Crysis 3 or Metro. If the game lags, graphical glitches begin to show up (artifacts such as miscolored polygons or pixels, 'snow', or anomalies on the textures) or anything you deem weird happens, you should back away to the last known safe value. Once you find the limits on stock voltage, you could being tweaking the voltage to get better results. Still, bear in mind that this is the part where you'll need to pay most attention to your temps, since voltage is what normally causes overheating problems. You can also tweak your fans to go up from stock if you find it's necessary.
If you're finding artifacts at a certain core or memory value, increasing the voltage should fix that. Be careful, though - Always try to do slight increases instead of just sliding the bar to it's limit and checking if it's working. Each overclocker has it's own 'stability' check - perhaps it's running a single benchmark, perhaps it's running a game 24/7, perhaps it's running a stress test such as Furmark. Only you'll be able to tell what's stable, but if your GPU passes a few runs of demanding benchmarks, it should be safe and stable.
Last but not least, whilst not absolutely safe, overclocking is safer when you do it with time and care. Unless you're into extreme OC'ing, if you do take your time and do things without rushing, chances you'll fry your card are small.