I have a stream of signal, and I'm looking for the presence of a certain event (the signal), also I have a "model" for that signal, and I will adopt it as my matched filter.

1. Could I cross-correlate the matched filter every unit of time (for example, every 2 seconds) and see if the value of cross correlation approaches to 1, and based on that decide if that event is existing or not?
2. What if my model, including a noise plus the signal, how much my works will be valid? 
3. What if I choose the matched filter as what is given in the literature (taking the matched filter as to increase the SNR) does that will increase the noise that my model is including it in addition to the signal? 

I think the hardest work is how to get the signal, not how to find the matched filter. Especially if it's biomedical signal.