As far as sexual selection is concerned, I like the view outlined by R. A. Fisher
In a nutshell, sexual selection begins when a female or group of females becomes genetically predisposed to find some characteristic of males attractive, meaning they are more likely to choose to mate with individuals with this characteristic.
(note: I'm using "female" and "male" here for illustrative purposes only. The two could be switched and it would make no difference to the basic idea.)
If a female has a mutation that makes her pick males with longer than average tail feathers, her offspring will carry both the gene for long tail feathers AND the gene for being attracted to long tail feathers. Basically, it creates a sub-population that will tend to mate only with members of that sub-population. Because they still tend to like longer feathers, they will keep selecting for longer and longer feathers. Given enough time, the sub-population either splits off or overruns the rest of the species.
Anyone familiar with the idea of positive feedback shouldn't be surprised that this could eventually lead to a modern peacock tail. And all that is required for the process to get started is a random push in one direction or the other.
The same principle can apply to antlers, plumage color or any number of observed traits.
Fisher actually modeled the whole thing mathematically, and contrary to popular belief, the trait DOES NOT need to be an indicator of general fitness for the model to work.