Prebuilt armv6 binaries + sample iOS App

• Code cleanup.
• C++11 fixes.
• Sample iOS App.
• Prebuilt armv6 binaries for easy demo with RPi zero + camera.
• Cotire cmake build.
• CUDA backend support for VexCL.

Working MLP+CONVNET Implementations

In addition to previous MLP implementations, this release contains two working CONVNET implementations:

• A single threaded implementation.
• A multi-threaded training cabable implementation.

Though not fully configurable yet, CONVNET implementations allow to choose amongst different state of the art solvers, activation functions and loss functions.

Neurocl now also contains a python wrapper for easy scripting, training data graph capabilities, sample mnist training app, as well as a digit ocr app.

Initial working MLP implementations

This releases contains three MLP implementations:

• VexCL based implementation.
• boost::ublas based implementation with standard operators.
• boost::ublas based fast implementation with SIMD optimized operators.