I haven't used neural networks, which is considered deep learning but I have built price algorithm using machine learning, specifically, supervised regression. Setting up an algorithm is not hard, the hard part would be making sure you have clean and normalized data and making sure your algorithm doesn't overfit your training data (can predict data it has seen very accurately but does not do a good job on data it hasn't seen).
All the tools I use are open source. Mainly python and numerical packages developed for use with python.
A. Gather Tools
1) Python
2) Numpy - numerical package for python
3) Pandas - dataframe
4) Sklearn - machine learning package http://scikit-learn.org/stable/documentation.html
5) Ta-Lib - * for techincal indicators if you are using them
B. Gather Data
If you're just looking for historical prices, you can use the datareader function in pandas to pull stock data from yahoo.
Ex.
from pandas_datareader import data
data = data.DataReader('AMZN','yahoo', '2009-01-01')
If you're looking for futures data, you can use quandl.
C. Clean Data
The most important part is cleaning the data. This means transforming, normalizing, and dropping bad data. Your model is only as good as the data it is being fed.
D. Train Your Data
This is where the machine learning parts come in. You will feed your algorithm features(data points that is used to predict your target) and it will try to find the best combination that represent the target. You want to use cross validation to split then train your data. This is just taking a random sample of your data, usually a 80/20 split, and splitting it into training data and testing data. Your algorithm will learn from the training data and then you will use the testing data to evaluate the performance of your model.