On larger devices, I tend to group them together by function. With smaller devices I might start out with the pins arranged as per the physical pinout, but I'll happily move them around if it's likely to make the resulting schematic easier to read.
With few exceptions, power and ground pins get separated out into another symbol. This keeps them well out of the way and keeps the signal flow uncluttered. They're *always* made visible and *always* explicitly connected to the correct power rail.
Many years ago I was working as a design engineer for a major networking company. They had a parts library which was centrally managed, and it was the job of a small team of component librarians to actually draw up symbols for any new parts the engineers wanted to use.
We had to submit the component's data sheet, get commercial approval to use the part (please allow several weeks), then wait for the symbol to be drawn up for us. Only then could it go on the schematic.
They had a rule for symbols: inputs go on the left, and outputs on the right. Just like we were taught at school, when the most complex device has about four pins.
So, you want to use a device with a crystal oscillator in it? Fine, XI is on the left, and XO on the right. Hmm, yes, that's clear and uncluttered.
An FPGA? No problem. The JTAG input pins are on the left, TDO is on the right, and all the bidirectional I/O pins are top or bottom.