Every character you add to the alphabet increases the time required to brute-force by 1/alphabet size of the original time (e.g. 1/26th)
Every normal character you add to the end of an existing password increases the time required to brute-force by the size of the alphabet (e.g. multiply by 26).
Say you use all the lower ASCII range (unlikely, lots of them aren't even printable, let alone acceptable in a password). That's 128 characters to the power of the length of your password.
Then say you use only the alphabet. That's 52 to the power of the length of a longer password.
128^8 = 72057594037927936
52^10 = 144555105949057024
An 8-character, all-symbol password takes half the time to guess than a 10-character, only alphabetical letters password.
In mathematical terms, the exponent here vastly outweighs the mantissa. n^m is much more affected by m as numbers increase than by n.
Put simply, the more characters you have, the more it swamps whatever silly symbols you decide to add to the alphabet. And, as stated, leet-speak is not even adding a whole character to the alphabet but just trying binary replacements of 'a' with '@' for example - it's even QUICKER to check for than just including @ as a random generic symbol, especially where dictionary attacks are concerned.
Stop messing about, and get a long, simple password.