Both approaches are valid, tho the resulting flexibility of usage is different.
Vector Graphic approach
You can use any vector design software to create a Enhanced Metafile (EMF). I would use Inkscape, but Adobe Illustrator and Corel Draw are able to create this files too. Remember to transform any Fonts in Outlines.
To use the emf file as a Marker Symbol, you need to choose Picture Marker Symbol Type in the Symbol Properties, and to browse the emf file.
Pros: Very simple processing;
Cons: You can not change the symbol colors in ArcMap, for that you need to create a new emf symbol.
Vector Graphic approach
Using Inkscape and the Online Font Converter you can follow this tutorial made by cleversomeday blog to produce a set of Fonts to use in ArcMap to create simple or composed marker symbols.
After installed In your computer you can use it as character marker symbol type.
If you need to create composed symbols, take special care with the size and alignment of the several fonts, so once in Arcmap they correctly overlap in ArcMap.
Pros: You can reuse the same symbol (font) with different color without changing the original file, making it more suitable to use in composed symbols;
Cons: It's a more complicated and time consuming process. It's difficult to create symbols with many different colors (ou need to create one font character per color)
How to decide what approach to use?
It depends of your needs and available time. If you are in a hurry to create a fast symbol or you know that you wont need any color variations from it, go for the EMF creation (Try to keep all your emf files in one folder, or you can very easily forget where is the original EMF file that you use in a certain project).
If in the other hand, If you have time and want to create a set of custom symbols to use in composed symbols, in many different projects and with infinite color and order combinations, you should go for the font creation.
Best Answer
Generally, ESRI symbols are fonts. So, just like with any font installed on your system the size is based on the character stored in the font. (Even north arrows are fonts, which is why they scale so well and they sometimes show up as letters in PDFs that you try to print on machines that don't have the ESRI fonts installed). The size of the symbol depends on whether you're looking at it in data or layout view. In data view, size is based on the current scale. In layout view, size is relative to the size of the document you are working on.
Here's a nice ESRI breakdown of how the fonts for symbols work: http://blogs.esri.com/esri/arcgis/2012/03/28/fonts-in-arcgis-symbols/
The major exception are symbols that are stored as graphic files, which is really a whole other subject. These symbols are based on individual image files and don't behave in the same ways that fonts do.