Actually, there is a utility that does exactly this. I know cos it was me wot wrote it. It was written for *BSD but ought to compile on Linux if you have the BSD libraries (which I believe are common).
I've just released a new version, posted here:
http://blog.frankleonhardt.com/2015/freebsd-hr-utility-human-readable-number-filter-man-page/
It's called hr, and it will take stdin (or files) and convert numbers to human-readable format in a way that is (now) exactly the same as ls -h and so on, and it can select individual feeds in lines, scale pre-scaled units (e.g. if they're in 512-byte blocks convert them to Mb etc), adjust column padding, and so on.
I wrote it a few years ago because I thought trying to write a shell script, although intellectually interesting, was also utter madness.
Using hr, for example, you can easily get a sorted list of directory sizes (which come out in 1Kb units and need shifting before converting) with the following:
du -d1 | sort -n | hr -sK
While du will produce -h output, sort won't sort by it. The addition of -h to existing utilities is a classic case of not following the unix philosophy: have simple utilities doing defined jobs really well.
