Why not Always Random Forest in Place of Linear or Logistic Regression Why Don't we use Random Forest always in place of Linear or Logistic Regressions. When will Linear and Logistic out perform Random Forest.  
 A: Advantages of linear models
1) They are much easier to interprete
2) You can do more than just predictions, e.g. test statistical hypotheses
3) They don't require much space to store
4) They are very fast to fit, depending on the chosen algorithm
When can linear models outperform a random forest?


*

*If you have enough time and knowledge to do smart feature construction/interaction selection etc. and the data set is not too large.

*...?
A: In addition to the points raised by @Michael M, we have to look at what exactly we mean by "outperform". 
One purpose of many statistical analyses is prediction. For this, I think RF might always be better than linear regression (I'm not sure about that, but it seems to be so, more knowledgeable users can bring up exceptions and details, Michael mentions one possibility; and, in a comment, @whuber mentions another).
But sometimes prediction is only part of our aim and sometimes it isn't even a part. Another goal of many analyses is explanation. Linear and logistic regression results explain the relationships among the independent and dependent variables in ways that may add to knowledge - or that may indicate some problem in the data or something to look into or whatnot.
In his wonderful book Statistics as Principled Argument Robert Abelson talks about another reason why regression might be better: It can be part of a principled argument in ways that would be difficult for random forests. 
A: Random Forest is good for sort of "catching" joint effects of 2 or more variables, meaning if feature 1 is above some threshold and feature for example below some other threshold, they will have different effect on target variable then you could model with linear relationship due to linear regression being additive model (you just sum up portion of effects of each feature). This means that random forest can outperform linear models in prediction, but since RF is ensamble method (comprised of many trees voting for the final prediction) it means you cannot interpret how each feature variable is affecting target variable. 
A: Random forest , Original learning dataset is randomly divided into several subsets of equal size, so its takes lot of memory and learning may be slow compared to linear and logistic regression. 
and Random forest do both classification and regression.
