Use of extern in in the implementation of double floats from an external library

I would say move up to the Arm processors that support 64-bit doubles directly. Some of them (like Teensy 3.0/3.1, Arduino Due, and Digistump DigiX) are programmed with variants of the Arduino IDE, so you can move most code over easily. Note, the Arm boards run at 3.3v instead of 5v, so you may need to deal with this with things that need 5v.

However note, all of the above chips emulate floating point. If you are doing a lot of FP calculations, I would recommend going up to the processors that run embedded Linux like Rasberry Pi, Beagle Bone Black, or pcDunio. These processors run at a higher clock rate, have more memory, have native floating point instructions. If you need to do real time processing, considering adding an Arduino as a slave processor to do the real time stuff.

Recently there was a kickstarter project for selling a Sparkcore that had a Spark V9 chip in it. The project has ended, but I don't know if it is now available via retail sales.