Andrew's provided you with clear definitions. Benchmarks depend on the area and they require that you have a term of comparison, and this can be a significant difficulty, requiring to gather data from several projects/organizations and eventually during several years. See for example: http://www.qsm.com/resources/qsm-database
About classifications, any classification is done based on some set of characteristics. Usually, this set is limited. I did propose a distinct approach here:
Also, I like to distinguish among classifications, but that is another issue.
Conference Paper Towards an Internal Numerical Taxonomy of Software Process A...
Emigdio, that's another question and the answer can be found inside the SWEBOK (see the answer above by James Moore). I only add that there is also a non free printed book about SWEBOK at amazon.