I just looked up the specs of that board, and I can tell you that if you're really limited to 512KB memory, you'll have a hard time running a neural network, and certainly one capable of character recognition!
I can speak of personal experience: 30 years ago I used what was then considered a powerful workstation, with 16 MB of RAM and processing speeds on the order of maybe 30-50 MHz. The processor you have may be faster than that, but the available memory is a far cry from what you'd need for character recognition.
Back then, we hand-crafted a very special neural network structure because we needed to preserve memory. Much later we've learned that others had done a similar thing, and named it
Convolutional neural network - Wikipedia[
^] . It 'only' required about 4MB RAM, leaving us with enough space to store training sets from about 20 different computer fonts. In the end we were able to recognize the majority of characters from a dozen other fonts that we've held back for testing. Note that all of these were computer fonts, not handwritten, and we did not require separate algorithms (and memory) for picking the outlines of single characters from the middle of a hand-written word.
It took the three of us several months to get this working in our free time (IIRC around 30 hours/week avg), and all of us already had experience with the design and programming of neural networks, C++ (not C!), and parallel programming.
Tl;dr: It's a helluva tough job to pull off, especially if you intend to do this on your own, without preexissting knowledge of the topic(s) at hand. And it's near impossible with less than 10 MB of RAM. Don't even mention handwritten characters.