Sport Statistical Modelling

I am very interested in football and want to create a statistical model for predicting the amount of goals scored by football team by using past results (goals scored, goals conceded etc). The problem I am facing is what sort of distribution represents these football results most realistically. I have enough data to assume that it will possibly be normally distributed however, regarding football results there comes a time limit when past results become less significant than more recent results. I would like some help to answer the following questions.

1) Should I use the normal or poisson distribution to calculate probabilities? Feel free to suggest another model.

2) Do you think that the normal distribution would represent football results well with a large sample? I was informed by someone on another forum that it would however I am doubtful of this as realistically I think there would be a positive skew of data (i.e. more teams scoring 0 than something towards the other end of the distribution such as 10).

3) How could I possibly add more weighting to the most recent results?