If I hadn't taken a Photoshop course in college, doing everything with a standard Paint-style program would probably not be worth the extra effort. Personally, I use Paint Shop Pro which is a similar program with most (if not all) the same features I need for this. Over time, I have accumulated a large number of map images, artwork, and homemade tokens and templates for my own personal "map-making studio" that I use for my online game. It still takes some work to finish, but the system I've developed makes it easy to manage.
The first thing that I did was create a generic numbered grid where each square is 50 x 50 pixels.
Whenever I decide on a particular map to use, I rescale it to get as close to the 50 pixel scale and paste it into the blank grid so the numbers and letters line up. Save it under a different file name and then crop the edges as needed. It doesn't have to be perfect, just good enough to differentiate C-4 and C-5, or any other space on the map.
I also have a growing library of custom-made tokens sized appropriately for my chosen scale of 50 pixels per square. As I prepare for an encounter, I select or create generic tokens for what I want. Then I'll make copies and add an alphanumeric designation specific for that encounter, such as L1, B2, etc. This makes it easier for players to designate their targets (i.e. "I attack Goblin (S2)"). I keep all the tokens for a specific encounter in the same folder on my drive so when it is time to update the map, I don't have to hunt them all down. It takes a little extra work when setting up, but the time saved doing continuous updates is worth the effort. Plus, all the efforts continue to grow my personal collection of tokens if I ever decide to run more games in the future, or just reuse old tokens for encounters with the same monsters.
There may be simpler ways, and others might see this as more work than its worth. Personally, I like the higher quality resulting from extra care and attention. But I wouldn't be opposed to something more intuitive and quicker if I could get the same results, and without a steep learning curve involved.