But that’s exactly what gzip does. You may think okay, it looks for repeated patterns, but then after a decompression the browser sees tons of excessive-looking

But that’s exactly what gzip does. You may think okay, it looks for repeated patterns, but then after a decompression the browser sees tons of excessive-looking styles, classes, etc. So what? They are just temporary strings pipelined from gunzip routine into a parser, which then caches similar patterns into same structures anyway. Doing it by hand would unlikely save any cpu time, memory, or network bandwidth, at all. A single word of lorem ipsum has more pixel edges to hint and antialias than there are bytes in your html+css. If you’re looking for a wider task, e.g. unused css class elimination, then start with webpack and its plugins.