So in fact, you can build a One-way function from a Collision Resistant Function.
Using a function family of CR functions F, we can create a one way function that works as follows:
pick function F from CRF family, F = ${f_k}$
$g =\ one\ way\ function$
$x_1\ and\ x_2$ picked from random
$g(x_1, x_2) = f_{x_1}(x_2),\ x_1$
The reason this works is because in order to break this one way function you would have to break the fact that $f_{x_1}$ is collision resistant.
Proof:
So let us assume there is an adversary A
that can invert CR function $f_{x_1}(x)\rightarrow y$.
$A(y) \rightarrow x$
What A returns is a pre-image of x so it can be the original x or a collision.
So now lets assume there exists an A' that can use A to break our one way function.
The way he will do so is by first allowing the random inputs of $x_1\ and\ x_2$.
$g(x_1, x_2)= f_{x_1}(x_2),\ x_1 \rightarrow y$
Now A' uses $A(y)\rightarrow x$ The output x is either = $x_1$ or a collision. With this A' has now successfully broken my scheme. But because we know that an adversary A that can find a collision/invert a collision resistant function does not exist in polynomial time we know our scheme is safe and one way because if one can not find a collision in polynomial time then one can not break the one wayness of my function in polynomial time.