Due to accuracy considerations, I'm using floating point in my application, but I am only using about 5 or so floating point data types or variables, the rest are fixed point.

Floating point isn't more or less accurate than a fixed point number of the same width. In fact, since the "exponent" is usually known and fixed in signal processing applications, a floating point variable has _less_ bits available for the mantissa than a fixed point variable.

Also, floating point variables bring all sort of nastiness with them, like their link between the absolute value and the granularity of a value (as a floating point variable grows larger, the delta between two adjacent numbers grows larger, too), floating point exceptions, odd rules for adding several floating point numbers together (do _you_ know how to correctly add three floating point numbers ?), and so on.

Floating point numbers are only "convenient" until one of these issues rears its ugly head and comes to bite you in the hindquarters.