What about boost string manipulation classes:
http://www.boost.org/libs/libraries.htm#String
Matrices ... I have tried loads of different libraries and all have benefits and drawbacks but I have yet to find something universally available (i.e. really common) yet simple and easy to use (such as Matlab and Python).
Boost, above, has a growing BLAS implementation but last time I looked things were still fairly spartan with regards to higher level manipulation, see
http://www.boost.org/libs/numeric/ublas/doc/index.htm - if you click on the Overview link at the top, they discuss some other efforts that have been made in this area (these might meet your needs too, IIRC, I had trouble building some of them on all platforms).
IMHO, uBLAS suffers (or did) from over-cryptification and documentation can be confusing. I think they have started a wiki now, so it might be better.
My own needs tend to be met either using a homegrown matrix / vector classes (i.e. simple transformations, limited and fixed functionality, fixed sizes) or building specific systems and using (C)LAPACK to solve them directly from c++ code, link
http://www.netlib.org/lapack/ . It will be somewhere in your distibution, you might want to peruse this link, too, to get you started, depending on your experience :
http://seehuhn.de/comp/linear.html
You could have a look at GSL which is more C than C++,
http://sources.redhat.com/gsl/ , but is fairly easy to use and a lot more straight forward than uBLAS, imho. IIRC, it didn't have all the functionality I needed, but that was sometime ago now.
All depends what you want to do with the matrices, really.